From c62a902b21dc2b520409509a952551f5ab8c32fc Mon Sep 17 00:00:00 2001 From: awkrail Date: Wed, 4 Sep 2024 15:42:39 +0900 Subject: [PATCH 1/8] add config_new --- configs_new/base.yml | 42 +++++++++++++++++++++++ configs_new/dataset/activitynet.yml | 2 ++ configs_new/dataset/charades.yml | 1 + configs_new/dataset/qvhighlight.yml | 2 ++ configs_new/dataset/tacos.yml | 4 +++ configs_new/dataset/tvsum.yml | 4 +++ configs_new/dataset/youtube_highlight.yml | 4 +++ configs_new/model/cg_detr.yml | 0 configs_new/model/eatr.yml | 0 configs_new/model/moment_detr.yml | 0 configs_new/model/qd_detr.yml | 0 configs_new/model/taskweave.yml | 0 configs_new/model/tr_detr.yml | 0 configs_new/model/uvcom.yml | 0 14 files changed, 59 insertions(+) create mode 100644 configs_new/base.yml create mode 100644 configs_new/dataset/activitynet.yml create mode 100644 configs_new/dataset/charades.yml create mode 100644 configs_new/dataset/qvhighlight.yml create mode 100644 configs_new/dataset/tacos.yml create mode 100644 configs_new/dataset/tvsum.yml create mode 100644 configs_new/dataset/youtube_highlight.yml create mode 100644 configs_new/model/cg_detr.yml create mode 100644 configs_new/model/eatr.yml create mode 100644 configs_new/model/moment_detr.yml create mode 100644 configs_new/model/qd_detr.yml create mode 100644 configs_new/model/taskweave.yml create mode 100644 configs_new/model/tr_detr.yml create mode 100644 configs_new/model/uvcom.yml diff --git a/configs_new/base.yml b/configs_new/base.yml new file mode 100644 index 0000000..a9799aa --- /dev/null +++ b/configs_new/base.yml @@ -0,0 +1,42 @@ +seed: 2023 +device: 'cuda' +num_workers: 4 + +lr: 0.0001 +lr_drop: 400 +wd: 0.0001 +n_epoch: 200 +max_es_cnt: 200 +bsz: 32 +eval_bsz: 100 +grad_clip: 0.1 +max_q_l: 32 +max_v_l: 75 +max_windows: 5 +clip_length: 1 +eval_epoch_interval: 1 + +position_embedding: "sine" +enc_layers: 2 +dec_layers: 2 +dim_feedforward: 1024 +hidden_dim: 256 +input_dropout: 0.5 +dropout: 0.1 +nheads: 8 +num_queries: 10 + +n_input_proj: 2 +saliency_margin: 0.2 +span_loss_type: "l1" +set_cost_span: 10 +set_cost_giou: 1 +set_cost_class: 4 +span_loss_coef: 10 +giou_loss_coef: 1 +label_loss_coef: 4 +eos_coef: 0.1 +lw_saliency: 1 + +model_ema: False +ema_decay: 0.9 \ No newline at end of file diff --git a/configs_new/dataset/activitynet.yml b/configs_new/dataset/activitynet.yml new file mode 100644 index 0000000..2b2458e --- /dev/null +++ b/configs_new/dataset/activitynet.yml @@ -0,0 +1,2 @@ +dset_name: activitynet +clip_length: 2 \ No newline at end of file diff --git a/configs_new/dataset/charades.yml b/configs_new/dataset/charades.yml new file mode 100644 index 0000000..befbdf5 --- /dev/null +++ b/configs_new/dataset/charades.yml @@ -0,0 +1 @@ +dset_name: charades \ No newline at end of file diff --git a/configs_new/dataset/qvhighlight.yml b/configs_new/dataset/qvhighlight.yml new file mode 100644 index 0000000..6c7b514 --- /dev/null +++ b/configs_new/dataset/qvhighlight.yml @@ -0,0 +1,2 @@ +dset_name: qvhighlight +clip_length: 2 \ No newline at end of file diff --git a/configs_new/dataset/tacos.yml b/configs_new/dataset/tacos.yml new file mode 100644 index 0000000..208f48c --- /dev/null +++ b/configs_new/dataset/tacos.yml @@ -0,0 +1,4 @@ +dset_name: tacos +clip_length: 2 +max_v_l: -1 +max_q_l: 100 \ No newline at end of file diff --git a/configs_new/dataset/tvsum.yml b/configs_new/dataset/tvsum.yml new file mode 100644 index 0000000..5bd7530 --- /dev/null +++ b/configs_new/dataset/tvsum.yml @@ -0,0 +1,4 @@ +dset_name: tvsum +clip_length: 2 +max_q_l: 100 +max_v_l: 1000 \ No newline at end of file diff --git a/configs_new/dataset/youtube_highlight.yml b/configs_new/dataset/youtube_highlight.yml new file mode 100644 index 0000000..36d6e97 --- /dev/null +++ b/configs_new/dataset/youtube_highlight.yml @@ -0,0 +1,4 @@ +dset_name: youtube_highlight +clip_length: 1 +max_q_l: 100 +max_v_l: 1000 \ No newline at end of file diff --git a/configs_new/model/cg_detr.yml b/configs_new/model/cg_detr.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/model/eatr.yml b/configs_new/model/eatr.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/model/moment_detr.yml b/configs_new/model/moment_detr.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/model/qd_detr.yml b/configs_new/model/qd_detr.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/model/taskweave.yml b/configs_new/model/taskweave.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/model/tr_detr.yml b/configs_new/model/tr_detr.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/model/uvcom.yml b/configs_new/model/uvcom.yml new file mode 100644 index 0000000..e69de29 From 34ba4f8109bbcf6b0209065db15d98e7678cd3fd Mon Sep 17 00:00:00 2001 From: awkrail Date: Wed, 4 Sep 2024 16:01:37 +0900 Subject: [PATCH 2/8] fix configs_new/dataset/ --- configs/base.yml | 3 +++ configs/qvhighlight/clip_slowfast_pann_uvcom_qvhighlight.yml | 1 + configs_new/dataset/activitynet.yml | 4 +++- configs_new/dataset/charades.yml | 4 +++- configs_new/dataset/qvhighlight.yml | 4 +++- configs_new/dataset/tacos.yml | 2 ++ configs_new/dataset/tvsum.yml | 2 ++ configs_new/dataset/youtube_highlight.yml | 2 ++ configs_new/feature/clip.yml | 0 configs_new/feature/clip_slowfast.yml | 0 configs_new/feature/clip_slowfast_pann.yml | 0 configs_new/feature/i3d_clip.yml | 0 configs_new/feature/resnet_glove.yml | 4 ++++ 13 files changed, 23 insertions(+), 3 deletions(-) create mode 100644 configs_new/feature/clip.yml create mode 100644 configs_new/feature/clip_slowfast.yml create mode 100644 configs_new/feature/clip_slowfast_pann.yml create mode 100644 configs_new/feature/i3d_clip.yml create mode 100644 configs_new/feature/resnet_glove.yml diff --git a/configs/base.yml b/configs/base.yml index 3c9f04a..0cab5cc 100644 --- a/configs/base.yml +++ b/configs/base.yml @@ -38,5 +38,8 @@ label_loss_coef: 4 eos_coef: 0.1 lw_saliency: 1 +train_log_filename: train.log +eval_log_filename: val.log + model_ema: False # If you use EMA, set True (Taskweave uses EMA) ema_decay: 0.9 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_pann_uvcom_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_pann_uvcom_qvhighlight.yml index d969487..4da8c93 100644 --- a/configs/qvhighlight/clip_slowfast_pann_uvcom_qvhighlight.yml +++ b/configs/qvhighlight/clip_slowfast_pann_uvcom_qvhighlight.yml @@ -2,6 +2,7 @@ model_name: uvcom dset_name: qvhighlight ctx_mode: video_audio_tef v_feat_types: slowfast_clip +a_feat_types: pann t_feat_type: clip train_path: data/qvhighlight/highlight_train_release.jsonl eval_path: data/qvhighlight/highlight_val_release.jsonl diff --git a/configs_new/dataset/activitynet.yml b/configs_new/dataset/activitynet.yml index 2b2458e..1d15eb8 100644 --- a/configs_new/dataset/activitynet.yml +++ b/configs_new/dataset/activitynet.yml @@ -1,2 +1,4 @@ dset_name: activitynet -clip_length: 2 \ No newline at end of file +clip_length: 2 +train_path: data/activitynet/activitynet_train_release.jsonl +eval_path: data/activitynet/activitynet_val_release.jsonl \ No newline at end of file diff --git a/configs_new/dataset/charades.yml b/configs_new/dataset/charades.yml index befbdf5..13eee53 100644 --- a/configs_new/dataset/charades.yml +++ b/configs_new/dataset/charades.yml @@ -1 +1,3 @@ -dset_name: charades \ No newline at end of file +dset_name: charades +train_path: data/charades/charades_train_release.jsonl +eval_path: data/charades/charades_test_release.jsonl \ No newline at end of file diff --git a/configs_new/dataset/qvhighlight.yml b/configs_new/dataset/qvhighlight.yml index 6c7b514..0b41532 100644 --- a/configs_new/dataset/qvhighlight.yml +++ b/configs_new/dataset/qvhighlight.yml @@ -1,2 +1,4 @@ dset_name: qvhighlight -clip_length: 2 \ No newline at end of file +clip_length: 2 +train_path: data/qvhighlight/highlight_train_release.jsonl +eval_path: data/qvhighlight/highlight_val_release.jsonl \ No newline at end of file diff --git a/configs_new/dataset/tacos.yml b/configs_new/dataset/tacos.yml index 208f48c..17dbabd 100644 --- a/configs_new/dataset/tacos.yml +++ b/configs_new/dataset/tacos.yml @@ -1,4 +1,6 @@ dset_name: tacos clip_length: 2 +train_path: data/tacos/tacos_train_release.jsonl +eval_path: data/tacos/tacos_test_release.jsonl max_v_l: -1 max_q_l: 100 \ No newline at end of file diff --git a/configs_new/dataset/tvsum.yml b/configs_new/dataset/tvsum.yml index 5bd7530..c59712f 100644 --- a/configs_new/dataset/tvsum.yml +++ b/configs_new/dataset/tvsum.yml @@ -1,4 +1,6 @@ dset_name: tvsum clip_length: 2 +train_path: data/tvsum/tvsum_train_release.jsonl +eval_path: data/tvsum/tvsum_val_release.jsonl max_q_l: 100 max_v_l: 1000 \ No newline at end of file diff --git a/configs_new/dataset/youtube_highlight.yml b/configs_new/dataset/youtube_highlight.yml index 36d6e97..e4e9827 100644 --- a/configs_new/dataset/youtube_highlight.yml +++ b/configs_new/dataset/youtube_highlight.yml @@ -1,4 +1,6 @@ dset_name: youtube_highlight clip_length: 1 +train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl +eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl max_q_l: 100 max_v_l: 1000 \ No newline at end of file diff --git a/configs_new/feature/clip.yml b/configs_new/feature/clip.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/feature/clip_slowfast.yml b/configs_new/feature/clip_slowfast.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/feature/clip_slowfast_pann.yml b/configs_new/feature/clip_slowfast_pann.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/feature/i3d_clip.yml b/configs_new/feature/i3d_clip.yml new file mode 100644 index 0000000..e69de29 diff --git a/configs_new/feature/resnet_glove.yml b/configs_new/feature/resnet_glove.yml new file mode 100644 index 0000000..ddd3b66 --- /dev/null +++ b/configs_new/feature/resnet_glove.yml @@ -0,0 +1,4 @@ +v_feat_types: resnet +t_feat_type: glove +v_feat_dim: 2050 +t_feat_dim: 300 \ No newline at end of file From c6a141d06d6c11697002de62286ffb689ca4a516 Mon Sep 17 00:00:00 2001 From: awkrail Date: Wed, 4 Sep 2024 18:54:18 +0900 Subject: [PATCH 3/8] checked config_news --- configs/base.yml | 3 --- configs_new/base.yml | 6 ++++++ configs_new/dataset/charades.yml | 6 +++++- configs_new/dataset/tacos.yml | 7 ++++++- configs_new/dataset/tvsum.yml | 7 ++++++- configs_new/dataset/youtube_highlight.yml | 8 +++++++- configs_new/feature/clip.yml | 5 +++++ configs_new/feature/clip_slowfast.yml | 5 +++++ configs_new/feature/clip_slowfast_pann.yml | 7 +++++++ configs_new/feature/i3d_clip.yml | 5 +++++ configs_new/feature/resnet_glove.yml | 1 + configs_new/model/cg_detr.yml | 14 ++++++++++++++ configs_new/model/eatr.yml | 2 ++ configs_new/model/moment_detr.yml | 1 + configs_new/model/qd_detr.yml | 1 + configs_new/model/taskweave.yml | 0 configs_new/model/taskweave_hd2mr.yml | 3 +++ configs_new/model/taskweave_mr2hd.yml | 3 +++ configs_new/model/tr_detr.yml | 3 +++ configs_new/model/uvcom.yml | 13 +++++++++++++ 20 files changed, 93 insertions(+), 7 deletions(-) delete mode 100644 configs_new/model/taskweave.yml create mode 100644 configs_new/model/taskweave_hd2mr.yml create mode 100644 configs_new/model/taskweave_mr2hd.yml diff --git a/configs/base.yml b/configs/base.yml index 0cab5cc..3c9f04a 100644 --- a/configs/base.yml +++ b/configs/base.yml @@ -38,8 +38,5 @@ label_loss_coef: 4 eos_coef: 0.1 lw_saliency: 1 -train_log_filename: train.log -eval_log_filename: val.log - model_ema: False # If you use EMA, set True (Taskweave uses EMA) ema_decay: 0.9 \ No newline at end of file diff --git a/configs_new/base.yml b/configs_new/base.yml index a9799aa..425d2a2 100644 --- a/configs_new/base.yml +++ b/configs_new/base.yml @@ -38,5 +38,11 @@ label_loss_coef: 4 eos_coef: 0.1 lw_saliency: 1 +ckpt_filename: best.ckpt +train_log_filename: train.log +eval_log_filename: val.log +eval_split_name: val +aux_loss: True + model_ema: False ema_decay: 0.9 \ No newline at end of file diff --git a/configs_new/dataset/charades.yml b/configs_new/dataset/charades.yml index 13eee53..bf93e6b 100644 --- a/configs_new/dataset/charades.yml +++ b/configs_new/dataset/charades.yml @@ -1,3 +1,7 @@ dset_name: charades train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl \ No newline at end of file +eval_path: data/charades/charades_test_release.jsonl + +lw_saliency: 4 +lr: 0.0002 +lr_drop: 40 \ No newline at end of file diff --git a/configs_new/dataset/tacos.yml b/configs_new/dataset/tacos.yml index 17dbabd..029f7db 100644 --- a/configs_new/dataset/tacos.yml +++ b/configs_new/dataset/tacos.yml @@ -3,4 +3,9 @@ clip_length: 2 train_path: data/tacos/tacos_train_release.jsonl eval_path: data/tacos/tacos_test_release.jsonl max_v_l: -1 -max_q_l: 100 \ No newline at end of file +max_q_l: 100 +eval_bsz: 32 + +lw_saliency: 4 +lr: 0.0002 +lr_drop: 40 \ No newline at end of file diff --git a/configs_new/dataset/tvsum.yml b/configs_new/dataset/tvsum.yml index c59712f..257650f 100644 --- a/configs_new/dataset/tvsum.yml +++ b/configs_new/dataset/tvsum.yml @@ -3,4 +3,9 @@ clip_length: 2 train_path: data/tvsum/tvsum_train_release.jsonl eval_path: data/tvsum/tvsum_val_release.jsonl max_q_l: 100 -max_v_l: 1000 \ No newline at end of file +max_v_l: 1000 +seed: 2018 +lr: 0.001 +lr_drop: 2000 +n_epoch: 2000 +bsz: 4 \ No newline at end of file diff --git a/configs_new/dataset/youtube_highlight.yml b/configs_new/dataset/youtube_highlight.yml index e4e9827..7aef189 100644 --- a/configs_new/dataset/youtube_highlight.yml +++ b/configs_new/dataset/youtube_highlight.yml @@ -3,4 +3,10 @@ clip_length: 1 train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl max_q_l: 100 -max_v_l: 1000 \ No newline at end of file +max_v_l: 1000 +bsz: 4 +seed: 2018 +lr: 0.0002 +lr_drop: 2000 +n_epoch: 1000 +lw_saliency: 4 \ No newline at end of file diff --git a/configs_new/feature/clip.yml b/configs_new/feature/clip.yml index e69de29..10929a1 100644 --- a/configs_new/feature/clip.yml +++ b/configs_new/feature/clip.yml @@ -0,0 +1,5 @@ +ctx_mode: video_tef +v_feat_types: clip +t_feat_type: clip +v_feat_dim: 514 +t_feat_dim: 512 \ No newline at end of file diff --git a/configs_new/feature/clip_slowfast.yml b/configs_new/feature/clip_slowfast.yml index e69de29..a6677f4 100644 --- a/configs_new/feature/clip_slowfast.yml +++ b/configs_new/feature/clip_slowfast.yml @@ -0,0 +1,5 @@ +ctx_mode: video_tef +v_feat_types: slowfast_clip +t_feat_type: clip +v_feat_dim: 2818 +t_feat_dim: 512 \ No newline at end of file diff --git a/configs_new/feature/clip_slowfast_pann.yml b/configs_new/feature/clip_slowfast_pann.yml index e69de29..0b105d7 100644 --- a/configs_new/feature/clip_slowfast_pann.yml +++ b/configs_new/feature/clip_slowfast_pann.yml @@ -0,0 +1,7 @@ +ctx_mode: video_audio_tef +v_feat_types: slowfast_clip +a_feat_types: pann +t_feat_type: clip +v_feat_dim: 2818 +a_feat_dim: 2048 +t_feat_dim: 512 \ No newline at end of file diff --git a/configs_new/feature/i3d_clip.yml b/configs_new/feature/i3d_clip.yml index e69de29..2cb0550 100644 --- a/configs_new/feature/i3d_clip.yml +++ b/configs_new/feature/i3d_clip.yml @@ -0,0 +1,5 @@ +ctx_mode: video_tef +v_feat_types: i3d +t_feat_type: clip +v_feat_dim: 2050 +t_feat_dim: 512 \ No newline at end of file diff --git a/configs_new/feature/resnet_glove.yml b/configs_new/feature/resnet_glove.yml index ddd3b66..e6d05f6 100644 --- a/configs_new/feature/resnet_glove.yml +++ b/configs_new/feature/resnet_glove.yml @@ -1,3 +1,4 @@ +ctx_mode: video_tef v_feat_types: resnet t_feat_type: glove v_feat_dim: 2050 diff --git a/configs_new/model/cg_detr.yml b/configs_new/model/cg_detr.yml index e69de29..c4ebc8c 100644 --- a/configs_new/model/cg_detr.yml +++ b/configs_new/model/cg_detr.yml @@ -0,0 +1,14 @@ +model_name: cg_detr +enc_layers: 3 +dec_layers: 3 +t2v_layers: 2 +sent_layers: 1 +moment_layers: 1 +dummy_layers: 2 +num_dummies: 45 +total_prompts: 10 +num_prompts: 2 +lw_saliency: 1 +lw_wattn: 1 +lw_ms_align: 1 +lw_distill: 1 \ No newline at end of file diff --git a/configs_new/model/eatr.yml b/configs_new/model/eatr.yml index e69de29..d4ec784 100644 --- a/configs_new/model/eatr.yml +++ b/configs_new/model/eatr.yml @@ -0,0 +1,2 @@ +model_name: eatr +event_coef: 3 \ No newline at end of file diff --git a/configs_new/model/moment_detr.yml b/configs_new/model/moment_detr.yml index e69de29..2b1164b 100644 --- a/configs_new/model/moment_detr.yml +++ b/configs_new/model/moment_detr.yml @@ -0,0 +1 @@ +model_name: moment_detr \ No newline at end of file diff --git a/configs_new/model/qd_detr.yml b/configs_new/model/qd_detr.yml index e69de29..1bc9af1 100644 --- a/configs_new/model/qd_detr.yml +++ b/configs_new/model/qd_detr.yml @@ -0,0 +1 @@ +model_name: qd_detr \ No newline at end of file diff --git a/configs_new/model/taskweave.yml b/configs_new/model/taskweave.yml deleted file mode 100644 index e69de29..0000000 diff --git a/configs_new/model/taskweave_hd2mr.yml b/configs_new/model/taskweave_hd2mr.yml new file mode 100644 index 0000000..bd7ae1d --- /dev/null +++ b/configs_new/model/taskweave_hd2mr.yml @@ -0,0 +1,3 @@ +model_name: taskweave +mr2hd: False +model_ema: True \ No newline at end of file diff --git a/configs_new/model/taskweave_mr2hd.yml b/configs_new/model/taskweave_mr2hd.yml new file mode 100644 index 0000000..a360b74 --- /dev/null +++ b/configs_new/model/taskweave_mr2hd.yml @@ -0,0 +1,3 @@ +model_name: taskweave +mr2hd: True +model_ema: True \ No newline at end of file diff --git a/configs_new/model/tr_detr.yml b/configs_new/model/tr_detr.yml index e69de29..e2dee6c 100644 --- a/configs_new/model/tr_detr.yml +++ b/configs_new/model/tr_detr.yml @@ -0,0 +1,3 @@ +model_name: tr_detr +VTC_loss_coef: 0.3 +CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs_new/model/uvcom.yml b/configs_new/model/uvcom.yml index e69de29..a2bfe95 100644 --- a/configs_new/model/uvcom.yml +++ b/configs_new/model/uvcom.yml @@ -0,0 +1,13 @@ +model_name: uvcom +enc_layers: 3 +dec_layers: 3 +num_queries: 30 +n_txt_mu: 5 +n_visual_mu: 30 +sim_loss_coef: 1 +neg_loss_coef: 0.5 +lr_gamma: 0.1 +neg_choose_epoch: 80 +lr_drop: 100 +em_iter: 5 +cross_fusion: False \ No newline at end of file From 629fbfe707d7683b7fa0a37c559eadcba9ac7139 Mon Sep 17 00:00:00 2001 From: awkrail Date: Thu, 5 Sep 2024 16:26:12 +0900 Subject: [PATCH 4/8] fix config file structure --- .../activitynet/clip_cg_detr_activitynet.yml | 34 ------ configs/activitynet/clip_eatr_activitynet.yml | 21 ---- .../clip_moment_detr_activitynet.yml | 18 ---- .../activitynet/clip_qd_detr_activitynet.yml | 18 ---- .../clip_slowfast_cg_detr_activitynet.yml | 34 ------ .../clip_slowfast_eatr_activitynet.yml | 21 ---- .../clip_slowfast_moment_detr_activitynet.yml | 18 ---- .../clip_slowfast_qd_detr_activitynet.yml | 18 ---- ...p_slowfast_taskweave_hd2mr_activitynet.yml | 23 ---- ...p_slowfast_taskweave_mr2hd_activitynet.yml | 23 ---- .../clip_slowfast_tr_detr_activitynet.yml | 22 ---- .../clip_slowfast_uvcom_activitynet.yml | 32 ------ .../clip_taskweave_hd2mr_activitynet.yml | 23 ---- .../clip_taskweave_mr2hd_activitynet.yml | 23 ---- .../activitynet/clip_tr_detr_activitynet.yml | 22 ---- .../activitynet/clip_uvcom_activitynet.yml | 32 ------ .../resnet_glove_cg_detr_activitynet.yml | 34 ------ .../resnet_glove_eatr_activitynet.yml | 21 ---- .../resnet_glove_moment_detr_activitynet.yml | 18 ---- .../resnet_glove_qd_detr_activitynet.yml | 18 ---- ...snet_glove_taskweave_hd2mr_activitynet.yml | 23 ---- ...snet_glove_taskweave_mr2hd_activitynet.yml | 23 ---- .../resnet_glove_tr_detr_activitynet.yml | 22 ---- .../resnet_glove_uvcom_activitynet.yml | 32 ------ configs/base.yml | 11 +- configs/charades/clip_cg_detr_charades.yml | 37 ------- configs/charades/clip_eatr_charades.yml | 25 ----- .../charades/clip_moment_detr_charades.yml | 22 ---- configs/charades/clip_qd_detr_charades.yml | 22 ---- .../clip_slowfast_cg_detr_charades.yml | 37 ------- .../charades/clip_slowfast_eatr_charades.yml | 25 ----- .../clip_slowfast_moment_detr_charades.yml | 22 ---- .../clip_slowfast_qd_detr_charades.yml | 22 ---- ...clip_slowfast_taskweave_hd2mr_charades.yml | 27 ----- ...clip_slowfast_taskweave_mr2hd_charades.yml | 27 ----- .../clip_slowfast_tr_detr_charades.yml | 26 ----- .../charades/clip_slowfast_uvcom_charades.yml | 36 ------- .../clip_taskweave_hd2mr_charades.yml | 27 ----- .../clip_taskweave_mr2hd_charades.yml | 27 ----- configs/charades/clip_tr_detr_charades.yml | 26 ----- configs/charades/clip_uvcom_charades.yml | 36 ------- .../resnet_glove_cg_detr_charades.yml | 37 ------- .../charades/resnet_glove_eatr_charades.yml | 25 ----- .../resnet_glove_moment_detr_charades.yml | 22 ---- .../resnet_glove_qd_detr_charades.yml | 22 ---- .../resnet_glove_taskweave_hd2mr_charades.yml | 27 ----- .../resnet_glove_taskweave_mr2hd_charades.yml | 27 ----- .../resnet_glove_tr_detr_charades.yml | 26 ----- .../charades/resnet_glove_uvcom_charades.yml | 36 ------- .../dataset/activitynet.yml | 0 {configs_new => configs}/dataset/charades.yml | 0 .../dataset/qvhighlight.yml | 0 {configs_new => configs}/dataset/tacos.yml | 0 {configs_new => configs}/dataset/tvsum.yml | 13 ++- .../dataset/youtube_highlight.yml | 9 +- {configs_new => configs}/feature/clip.yml | 0 .../feature/clip_slowfast.yml | 0 .../feature/clip_slowfast_pann.yml | 0 {configs_new => configs}/feature/i3d_clip.yml | 0 .../feature/resnet_glove.yml | 0 {configs_new => configs}/model/cg_detr.yml | 0 {configs_new => configs}/model/eatr.yml | 0 .../model/moment_detr.yml | 0 {configs_new => configs}/model/qd_detr.yml | 0 .../model/taskweave_hd2mr.yml | 0 .../model/taskweave_mr2hd.yml | 0 {configs_new => configs}/model/tr_detr.yml | 0 {configs_new => configs}/model/uvcom.yml | 0 .../qvhighlight/clip_cg_detr_qvhighlight.yml | 34 ------ configs/qvhighlight/clip_eatr_qvhighlight.yml | 21 ---- .../clip_moment_detr_qvhighlight.yml | 18 ---- .../qvhighlight/clip_qd_detr_qvhighlight.yml | 18 ---- .../clip_slowfast_cg_detr_qvhighlight.yml | 34 ------ .../clip_slowfast_eatr_qvhighlight.yml | 21 ---- .../clip_slowfast_moment_detr_qvhighlight.yml | 18 ---- ...wfast_moment_detr_qvhighlight_pretrain.yml | 19 ---- ...clip_slowfast_pann_cg_detr_qvhighlight.yml | 37 ------- .../clip_slowfast_pann_eatr_qvhighlight.yml | 24 ----- ..._slowfast_pann_moment_detr_qvhighlight.yml | 21 ---- ...clip_slowfast_pann_qd_detr_qvhighlight.yml | 21 ---- ...wfast_pann_taskweave_hd2mr_qvhighlight.yml | 26 ----- ...wfast_pann_taskweave_mr2hd_qvhighlight.yml | 26 ----- ...clip_slowfast_pann_tr_detr_qvhighlight.yml | 25 ----- .../clip_slowfast_pann_uvcom_qvhighlight.yml | 35 ------ .../clip_slowfast_qd_detr_qvhighlight.yml | 18 ---- ...p_slowfast_taskweave_hd2mr_qvhighlight.yml | 23 ---- ...p_slowfast_taskweave_mr2hd_qvhighlight.yml | 23 ---- .../clip_slowfast_tr_detr_qvhighlight.yml | 22 ---- .../clip_slowfast_uvcom_qvhighlight.yml | 32 ------ .../clip_taskweave_hd2mr_qvhighlight.yml | 23 ---- .../clip_taskweave_mr2hd_qvhighlight.yml | 23 ---- .../qvhighlight/clip_tr_detr_qvhighlight.yml | 22 ---- .../qvhighlight/clip_uvcom_qvhighlight.yml | 32 ------ .../resnet_glove_cg_detr_qvhighlight.yml | 34 ------ .../resnet_glove_eatr_qvhighlight.yml | 21 ---- .../resnet_glove_moment_detr_qvhighlight.yml | 18 ---- .../resnet_glove_qd_detr_qvhighlight.yml | 18 ---- ...snet_glove_taskweave_hd2mr_qvhighlight.yml | 23 ---- ...snet_glove_taskweave_mr2hd_qvhighlight.yml | 23 ---- .../resnet_glove_tr_detr_qvhighlight.yml | 22 ---- .../resnet_glove_uvcom_qvhighlight.yml | 32 ------ configs/tacos/clip_cg_detr_tacos.yml | 41 ------- configs/tacos/clip_eatr_tacos.yml | 29 ----- configs/tacos/clip_moment_detr_tacos.yml | 26 ----- configs/tacos/clip_qd_detr_tacos.yml | 26 ----- configs/tacos/clip_slowfast_cg_detr_tacos.yml | 41 ------- configs/tacos/clip_slowfast_eatr_tacos.yml | 29 ----- .../tacos/clip_slowfast_moment_detr_tacos.yml | 26 ----- configs/tacos/clip_slowfast_qd_detr_tacos.yml | 26 ----- .../clip_slowfast_taskweave_hd2mr_tacos.yml | 31 ------ .../clip_slowfast_taskweave_mr2hd_tacos.yml | 31 ------ configs/tacos/clip_slowfast_tr_detr_tacos.yml | 30 ------ configs/tacos/clip_slowfast_uvcom_tacos.yml | 40 ------- configs/tacos/clip_taskweave_hd2mr_tacos.yml | 31 ------ configs/tacos/clip_taskweave_mr2hd_tacos.yml | 31 ------ configs/tacos/clip_tr_detr_tacos.yml | 30 ------ configs/tacos/clip_uvcom_tacos.yml | 40 ------- configs/tacos/resnet_glove_cg_detr_tacos.yml | 41 ------- configs/tacos/resnet_glove_eatr_tacos.yml | 29 ----- .../tacos/resnet_glove_moment_detr_tacos.yml | 26 ----- configs/tacos/resnet_glove_qd_detr_tacos.yml | 26 ----- .../resnet_glove_taskweave_hd2mr_tacos.yml | 31 ------ .../resnet_glove_taskweave_mr2hd_tacos.yml | 31 ------ configs/tacos/resnet_glove_tr_detr_tacos.yml | 30 ------ configs/tacos/resnet_glove_uvcom_tacos.yml | 40 ------- configs/tvsum/clip_cgdetr_tvsum.yml | 43 -------- configs/tvsum/clip_eatr_tvsum.yml | 29 ----- configs/tvsum/clip_moment_detr_tvsum.yml | 26 ----- configs/tvsum/clip_qd_detr_tvsum.yml | 26 ----- configs/tvsum/clip_slowfast_cgdetr_tvsum.yml | 43 -------- configs/tvsum/clip_slowfast_eatr_tvsum.yml | 29 ----- .../tvsum/clip_slowfast_moment_detr_tvsum.yml | 26 ----- configs/tvsum/clip_slowfast_qd_detr_tvsum.yml | 26 ----- .../clip_slowfast_taskweave_hd2mr_tvsum.yml | 31 ------ .../clip_slowfast_taskweave_mr2hd_tvsum.yml | 31 ------ configs/tvsum/clip_slowfast_uvcom_tvsum.yml | 39 ------- configs/tvsum/clip_taskweave_hd2mr_tvsum.yml | 31 ------ configs/tvsum/clip_taskweave_mr2hd_tvsum.yml | 31 ------ configs/tvsum/clip_uvcom_tvsum.yml | 39 ------- configs/tvsum/i3d_clip_cgdetr_tvsum.yml | 43 -------- configs/tvsum/i3d_clip_eatr_tvsum.yml | 29 ----- configs/tvsum/i3d_clip_moment_detr_tvsum.yml | 26 ----- configs/tvsum/i3d_clip_qd_detr_tvsum.yml | 26 ----- .../tvsum/i3d_clip_taskweave_hd2mr_tvsum.yml | 31 ------ .../tvsum/i3d_clip_taskweave_mr2hd_tvsum.yml | 31 ------ configs/tvsum/i3d_clip_tr_detr_tvsum.yml | 30 ------ configs/tvsum/i3d_clip_uvcom_tvsum.yml | 39 ------- configs/tvsum/resnet_glove_cgdetr_tvsum.yml | 43 -------- configs/tvsum/resnet_glove_eatr_tvsum.yml | 29 ----- .../tvsum/resnet_glove_moment_detr_tvsum.yml | 26 ----- configs/tvsum/resnet_glove_qd_detr_tvsum.yml | 26 ----- .../resnet_glove_taskweave_hd2mr_tvsum.yml | 31 ------ .../resnet_glove_taskweave_mr2hd_tvsum.yml | 31 ------ configs/tvsum/resnet_glove_uvcom_tvsum.yml | 39 ------- .../clip_cgdetr_youtube_highlight.yml | 43 -------- .../clip_eatr_youtube_highlight.yml | 30 ------ .../clip_moment_detr_youtube_highlight.yml | 27 ----- .../clip_qd_detr_youtube_highlight.yml | 27 ----- ...clip_slowfast_cgdetr_youtube_highlight.yml | 43 -------- .../clip_slowfast_eatr_youtube_highlight.yml | 30 ------ ...slowfast_moment_detr_youtube_highlight.yml | 27 ----- ...lip_slowfast_qd_detr_youtube_highlight.yml | 27 ----- ...fast_taskweave_hd2mr_youtube_highlight.yml | 32 ------ ...fast_taskweave_mr2hd_youtube_highlight.yml | 32 ------ .../clip_slowfast_uvcom_youtube_highlight.yml | 40 ------- ...clip_taskweave_hd2mr_youtube_highlight.yml | 32 ------ ...clip_taskweave_mr2hd_youtube_highlight.yml | 32 ------ .../clip_uvcom_youtube_highlight.yml | 40 ------- configs_new/base.yml | 48 --------- training/config.py | 100 +++++++++++++----- training/{ => standalone_eval}/__init__.py | 0 training/train.py | 80 ++++++++++---- 172 files changed, 161 insertions(+), 4300 deletions(-) delete mode 100644 configs/activitynet/clip_cg_detr_activitynet.yml delete mode 100644 configs/activitynet/clip_eatr_activitynet.yml delete mode 100644 configs/activitynet/clip_moment_detr_activitynet.yml delete mode 100644 configs/activitynet/clip_qd_detr_activitynet.yml delete mode 100644 configs/activitynet/clip_slowfast_cg_detr_activitynet.yml delete mode 100644 configs/activitynet/clip_slowfast_eatr_activitynet.yml delete mode 100644 configs/activitynet/clip_slowfast_moment_detr_activitynet.yml delete mode 100644 configs/activitynet/clip_slowfast_qd_detr_activitynet.yml delete mode 100644 configs/activitynet/clip_slowfast_taskweave_hd2mr_activitynet.yml delete mode 100644 configs/activitynet/clip_slowfast_taskweave_mr2hd_activitynet.yml delete mode 100644 configs/activitynet/clip_slowfast_tr_detr_activitynet.yml delete mode 100644 configs/activitynet/clip_slowfast_uvcom_activitynet.yml delete mode 100644 configs/activitynet/clip_taskweave_hd2mr_activitynet.yml delete mode 100644 configs/activitynet/clip_taskweave_mr2hd_activitynet.yml delete mode 100644 configs/activitynet/clip_tr_detr_activitynet.yml delete mode 100644 configs/activitynet/clip_uvcom_activitynet.yml delete mode 100644 configs/activitynet/resnet_glove_cg_detr_activitynet.yml delete mode 100644 configs/activitynet/resnet_glove_eatr_activitynet.yml delete mode 100644 configs/activitynet/resnet_glove_moment_detr_activitynet.yml delete mode 100644 configs/activitynet/resnet_glove_qd_detr_activitynet.yml delete mode 100644 configs/activitynet/resnet_glove_taskweave_hd2mr_activitynet.yml delete mode 100644 configs/activitynet/resnet_glove_taskweave_mr2hd_activitynet.yml delete mode 100644 configs/activitynet/resnet_glove_tr_detr_activitynet.yml delete mode 100644 configs/activitynet/resnet_glove_uvcom_activitynet.yml delete mode 100644 configs/charades/clip_cg_detr_charades.yml delete mode 100644 configs/charades/clip_eatr_charades.yml delete mode 100644 configs/charades/clip_moment_detr_charades.yml delete mode 100644 configs/charades/clip_qd_detr_charades.yml delete mode 100644 configs/charades/clip_slowfast_cg_detr_charades.yml delete mode 100644 configs/charades/clip_slowfast_eatr_charades.yml delete mode 100644 configs/charades/clip_slowfast_moment_detr_charades.yml delete mode 100644 configs/charades/clip_slowfast_qd_detr_charades.yml delete mode 100644 configs/charades/clip_slowfast_taskweave_hd2mr_charades.yml delete mode 100644 configs/charades/clip_slowfast_taskweave_mr2hd_charades.yml delete mode 100644 configs/charades/clip_slowfast_tr_detr_charades.yml delete mode 100644 configs/charades/clip_slowfast_uvcom_charades.yml delete mode 100644 configs/charades/clip_taskweave_hd2mr_charades.yml delete mode 100644 configs/charades/clip_taskweave_mr2hd_charades.yml delete mode 100644 configs/charades/clip_tr_detr_charades.yml delete mode 100644 configs/charades/clip_uvcom_charades.yml delete mode 100644 configs/charades/resnet_glove_cg_detr_charades.yml delete mode 100644 configs/charades/resnet_glove_eatr_charades.yml delete mode 100644 configs/charades/resnet_glove_moment_detr_charades.yml delete mode 100644 configs/charades/resnet_glove_qd_detr_charades.yml delete mode 100644 configs/charades/resnet_glove_taskweave_hd2mr_charades.yml delete mode 100644 configs/charades/resnet_glove_taskweave_mr2hd_charades.yml delete mode 100644 configs/charades/resnet_glove_tr_detr_charades.yml delete mode 100644 configs/charades/resnet_glove_uvcom_charades.yml rename {configs_new => configs}/dataset/activitynet.yml (100%) rename {configs_new => configs}/dataset/charades.yml (100%) rename {configs_new => configs}/dataset/qvhighlight.yml (100%) rename {configs_new => configs}/dataset/tacos.yml (100%) rename {configs_new => configs}/dataset/tvsum.yml (70%) rename {configs_new => configs}/dataset/youtube_highlight.yml (74%) rename {configs_new => configs}/feature/clip.yml (100%) rename {configs_new => configs}/feature/clip_slowfast.yml (100%) rename {configs_new => configs}/feature/clip_slowfast_pann.yml (100%) rename {configs_new => configs}/feature/i3d_clip.yml (100%) rename {configs_new => configs}/feature/resnet_glove.yml (100%) rename {configs_new => configs}/model/cg_detr.yml (100%) rename {configs_new => configs}/model/eatr.yml (100%) rename {configs_new => configs}/model/moment_detr.yml (100%) rename {configs_new => configs}/model/qd_detr.yml (100%) rename {configs_new => configs}/model/taskweave_hd2mr.yml (100%) rename {configs_new => configs}/model/taskweave_mr2hd.yml (100%) rename {configs_new => configs}/model/tr_detr.yml (100%) rename {configs_new => configs}/model/uvcom.yml (100%) delete mode 100644 configs/qvhighlight/clip_cg_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_eatr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_moment_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_qd_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_cg_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_eatr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_moment_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_moment_detr_qvhighlight_pretrain.yml delete mode 100644 configs/qvhighlight/clip_slowfast_pann_cg_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_pann_eatr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_pann_moment_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_pann_qd_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_pann_taskweave_hd2mr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_pann_taskweave_mr2hd_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_pann_tr_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_pann_uvcom_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_qd_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_taskweave_hd2mr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_taskweave_mr2hd_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_tr_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_slowfast_uvcom_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_taskweave_hd2mr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_taskweave_mr2hd_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_tr_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/clip_uvcom_qvhighlight.yml delete mode 100644 configs/qvhighlight/resnet_glove_cg_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/resnet_glove_eatr_qvhighlight.yml delete mode 100644 configs/qvhighlight/resnet_glove_moment_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/resnet_glove_qd_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/resnet_glove_taskweave_hd2mr_qvhighlight.yml delete mode 100644 configs/qvhighlight/resnet_glove_taskweave_mr2hd_qvhighlight.yml delete mode 100644 configs/qvhighlight/resnet_glove_tr_detr_qvhighlight.yml delete mode 100644 configs/qvhighlight/resnet_glove_uvcom_qvhighlight.yml delete mode 100644 configs/tacos/clip_cg_detr_tacos.yml delete mode 100644 configs/tacos/clip_eatr_tacos.yml delete mode 100644 configs/tacos/clip_moment_detr_tacos.yml delete mode 100644 configs/tacos/clip_qd_detr_tacos.yml delete mode 100644 configs/tacos/clip_slowfast_cg_detr_tacos.yml delete mode 100644 configs/tacos/clip_slowfast_eatr_tacos.yml delete mode 100644 configs/tacos/clip_slowfast_moment_detr_tacos.yml delete mode 100644 configs/tacos/clip_slowfast_qd_detr_tacos.yml delete mode 100644 configs/tacos/clip_slowfast_taskweave_hd2mr_tacos.yml delete mode 100644 configs/tacos/clip_slowfast_taskweave_mr2hd_tacos.yml delete mode 100644 configs/tacos/clip_slowfast_tr_detr_tacos.yml delete mode 100644 configs/tacos/clip_slowfast_uvcom_tacos.yml delete mode 100644 configs/tacos/clip_taskweave_hd2mr_tacos.yml delete mode 100644 configs/tacos/clip_taskweave_mr2hd_tacos.yml delete mode 100644 configs/tacos/clip_tr_detr_tacos.yml delete mode 100644 configs/tacos/clip_uvcom_tacos.yml delete mode 100644 configs/tacos/resnet_glove_cg_detr_tacos.yml delete mode 100644 configs/tacos/resnet_glove_eatr_tacos.yml delete mode 100644 configs/tacos/resnet_glove_moment_detr_tacos.yml delete mode 100644 configs/tacos/resnet_glove_qd_detr_tacos.yml delete mode 100644 configs/tacos/resnet_glove_taskweave_hd2mr_tacos.yml delete mode 100644 configs/tacos/resnet_glove_taskweave_mr2hd_tacos.yml delete mode 100644 configs/tacos/resnet_glove_tr_detr_tacos.yml delete mode 100644 configs/tacos/resnet_glove_uvcom_tacos.yml delete mode 100644 configs/tvsum/clip_cgdetr_tvsum.yml delete mode 100644 configs/tvsum/clip_eatr_tvsum.yml delete mode 100644 configs/tvsum/clip_moment_detr_tvsum.yml delete mode 100644 configs/tvsum/clip_qd_detr_tvsum.yml delete mode 100644 configs/tvsum/clip_slowfast_cgdetr_tvsum.yml delete mode 100644 configs/tvsum/clip_slowfast_eatr_tvsum.yml delete mode 100644 configs/tvsum/clip_slowfast_moment_detr_tvsum.yml delete mode 100644 configs/tvsum/clip_slowfast_qd_detr_tvsum.yml delete mode 100644 configs/tvsum/clip_slowfast_taskweave_hd2mr_tvsum.yml delete mode 100644 configs/tvsum/clip_slowfast_taskweave_mr2hd_tvsum.yml delete mode 100644 configs/tvsum/clip_slowfast_uvcom_tvsum.yml delete mode 100644 configs/tvsum/clip_taskweave_hd2mr_tvsum.yml delete mode 100644 configs/tvsum/clip_taskweave_mr2hd_tvsum.yml delete mode 100644 configs/tvsum/clip_uvcom_tvsum.yml delete mode 100644 configs/tvsum/i3d_clip_cgdetr_tvsum.yml delete mode 100644 configs/tvsum/i3d_clip_eatr_tvsum.yml delete mode 100644 configs/tvsum/i3d_clip_moment_detr_tvsum.yml delete mode 100644 configs/tvsum/i3d_clip_qd_detr_tvsum.yml delete mode 100644 configs/tvsum/i3d_clip_taskweave_hd2mr_tvsum.yml delete mode 100644 configs/tvsum/i3d_clip_taskweave_mr2hd_tvsum.yml delete mode 100644 configs/tvsum/i3d_clip_tr_detr_tvsum.yml delete mode 100644 configs/tvsum/i3d_clip_uvcom_tvsum.yml delete mode 100644 configs/tvsum/resnet_glove_cgdetr_tvsum.yml delete mode 100644 configs/tvsum/resnet_glove_eatr_tvsum.yml delete mode 100644 configs/tvsum/resnet_glove_moment_detr_tvsum.yml delete mode 100644 configs/tvsum/resnet_glove_qd_detr_tvsum.yml delete mode 100644 configs/tvsum/resnet_glove_taskweave_hd2mr_tvsum.yml delete mode 100644 configs/tvsum/resnet_glove_taskweave_mr2hd_tvsum.yml delete mode 100644 configs/tvsum/resnet_glove_uvcom_tvsum.yml delete mode 100644 configs/youtube_highlight/clip_cgdetr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_eatr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_moment_detr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_qd_detr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_slowfast_cgdetr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_slowfast_eatr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_slowfast_moment_detr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_slowfast_qd_detr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_slowfast_taskweave_hd2mr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_slowfast_taskweave_mr2hd_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_slowfast_uvcom_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_taskweave_hd2mr_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_taskweave_mr2hd_youtube_highlight.yml delete mode 100644 configs/youtube_highlight/clip_uvcom_youtube_highlight.yml delete mode 100644 configs_new/base.yml rename training/{ => standalone_eval}/__init__.py (100%) diff --git a/configs/activitynet/clip_cg_detr_activitynet.yml b/configs/activitynet/clip_cg_detr_activitynet.yml deleted file mode 100644 index 4ba5606..0000000 --- a/configs/activitynet/clip_cg_detr_activitynet.yml +++ /dev/null @@ -1,34 +0,0 @@ -model_name: cg_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_cg_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/activitynet/clip_eatr_activitynet.yml b/configs/activitynet/clip_eatr_activitynet.yml deleted file mode 100644 index 599907e..0000000 --- a/configs/activitynet/clip_eatr_activitynet.yml +++ /dev/null @@ -1,21 +0,0 @@ -model_name: eatr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_eatr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/activitynet/clip_moment_detr_activitynet.yml b/configs/activitynet/clip_moment_detr_activitynet.yml deleted file mode 100644 index 0c84a18..0000000 --- a/configs/activitynet/clip_moment_detr_activitynet.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: moment_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_moment_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/activitynet/clip_qd_detr_activitynet.yml b/configs/activitynet/clip_qd_detr_activitynet.yml deleted file mode 100644 index e897821..0000000 --- a/configs/activitynet/clip_qd_detr_activitynet.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: qd_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_qd_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/activitynet/clip_slowfast_cg_detr_activitynet.yml b/configs/activitynet/clip_slowfast_cg_detr_activitynet.yml deleted file mode 100644 index 5ab65e2..0000000 --- a/configs/activitynet/clip_slowfast_cg_detr_activitynet.yml +++ /dev/null @@ -1,34 +0,0 @@ -model_name: cg_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip', 'features/ActivityNet/slowfast'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_cg_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/activitynet/clip_slowfast_eatr_activitynet.yml b/configs/activitynet/clip_slowfast_eatr_activitynet.yml deleted file mode 100644 index 62d509a..0000000 --- a/configs/activitynet/clip_slowfast_eatr_activitynet.yml +++ /dev/null @@ -1,21 +0,0 @@ -model_name: eatr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip', 'features/ActivityNet/slowfast'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_eatr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/activitynet/clip_slowfast_moment_detr_activitynet.yml b/configs/activitynet/clip_slowfast_moment_detr_activitynet.yml deleted file mode 100644 index 85aab60..0000000 --- a/configs/activitynet/clip_slowfast_moment_detr_activitynet.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: moment_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip', 'features/ActivityNet/slowfast'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_moment_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/activitynet/clip_slowfast_qd_detr_activitynet.yml b/configs/activitynet/clip_slowfast_qd_detr_activitynet.yml deleted file mode 100644 index ccd720f..0000000 --- a/configs/activitynet/clip_slowfast_qd_detr_activitynet.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: qd_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip', 'features/ActivityNet/slowfast'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_qd_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/activitynet/clip_slowfast_taskweave_hd2mr_activitynet.yml b/configs/activitynet/clip_slowfast_taskweave_hd2mr_activitynet.yml deleted file mode 100644 index d5a5351..0000000 --- a/configs/activitynet/clip_slowfast_taskweave_hd2mr_activitynet.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip', 'features/ActivityNet/slowfast'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_hd2mr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/activitynet/clip_slowfast_taskweave_mr2hd_activitynet.yml b/configs/activitynet/clip_slowfast_taskweave_mr2hd_activitynet.yml deleted file mode 100644 index 3675370..0000000 --- a/configs/activitynet/clip_slowfast_taskweave_mr2hd_activitynet.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip', 'features/ActivityNet/slowfast'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_mr2hd/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/activitynet/clip_slowfast_tr_detr_activitynet.yml b/configs/activitynet/clip_slowfast_tr_detr_activitynet.yml deleted file mode 100644 index 4d2d48b..0000000 --- a/configs/activitynet/clip_slowfast_tr_detr_activitynet.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: tr_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip', 'features/ActivityNet/slowfast'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_tr_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/activitynet/clip_slowfast_uvcom_activitynet.yml b/configs/activitynet/clip_slowfast_uvcom_activitynet.yml deleted file mode 100644 index 7004a59..0000000 --- a/configs/activitynet/clip_slowfast_uvcom_activitynet.yml +++ /dev/null @@ -1,32 +0,0 @@ -model_name: uvcom -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip', 'features/ActivityNet/slowfast'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_uvcom/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/activitynet/clip_taskweave_hd2mr_activitynet.yml b/configs/activitynet/clip_taskweave_hd2mr_activitynet.yml deleted file mode 100644 index cd5af82..0000000 --- a/configs/activitynet/clip_taskweave_hd2mr_activitynet.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_hd2mr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/activitynet/clip_taskweave_mr2hd_activitynet.yml b/configs/activitynet/clip_taskweave_mr2hd_activitynet.yml deleted file mode 100644 index 352618b..0000000 --- a/configs/activitynet/clip_taskweave_mr2hd_activitynet.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_mr2hd/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/activitynet/clip_tr_detr_activitynet.yml b/configs/activitynet/clip_tr_detr_activitynet.yml deleted file mode 100644 index 1e75f26..0000000 --- a/configs/activitynet/clip_tr_detr_activitynet.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: tr_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_tr_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/activitynet/clip_uvcom_activitynet.yml b/configs/activitynet/clip_uvcom_activitynet.yml deleted file mode 100644 index 920fdf9..0000000 --- a/configs/activitynet/clip_uvcom_activitynet.yml +++ /dev/null @@ -1,32 +0,0 @@ -model_name: uvcom -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/clip'] -t_feat_dir: features/ActivityNet/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_uvcom/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/activitynet/resnet_glove_cg_detr_activitynet.yml b/configs/activitynet/resnet_glove_cg_detr_activitynet.yml deleted file mode 100644 index 553a658..0000000 --- a/configs/activitynet/resnet_glove_cg_detr_activitynet.yml +++ /dev/null @@ -1,34 +0,0 @@ -model_name: cg_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/resnet'] -t_feat_dir: features/ActivityNet/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_cg_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/activitynet/resnet_glove_eatr_activitynet.yml b/configs/activitynet/resnet_glove_eatr_activitynet.yml deleted file mode 100644 index 18d5a6f..0000000 --- a/configs/activitynet/resnet_glove_eatr_activitynet.yml +++ /dev/null @@ -1,21 +0,0 @@ -model_name: eatr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/resnet'] -t_feat_dir: features/ActivityNet/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_eatr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/activitynet/resnet_glove_moment_detr_activitynet.yml b/configs/activitynet/resnet_glove_moment_detr_activitynet.yml deleted file mode 100644 index 42d3203..0000000 --- a/configs/activitynet/resnet_glove_moment_detr_activitynet.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: moment_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/resnet'] -t_feat_dir: features/ActivityNet/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_moment_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/activitynet/resnet_glove_qd_detr_activitynet.yml b/configs/activitynet/resnet_glove_qd_detr_activitynet.yml deleted file mode 100644 index de7fccc..0000000 --- a/configs/activitynet/resnet_glove_qd_detr_activitynet.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: qd_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/resnet'] -t_feat_dir: features/ActivityNet/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_qd_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/activitynet/resnet_glove_taskweave_hd2mr_activitynet.yml b/configs/activitynet/resnet_glove_taskweave_hd2mr_activitynet.yml deleted file mode 100644 index 49f8fd9..0000000 --- a/configs/activitynet/resnet_glove_taskweave_hd2mr_activitynet.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/resnet'] -t_feat_dir: features/ActivityNet/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_hd2mr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/activitynet/resnet_glove_taskweave_mr2hd_activitynet.yml b/configs/activitynet/resnet_glove_taskweave_mr2hd_activitynet.yml deleted file mode 100644 index 1585b9f..0000000 --- a/configs/activitynet/resnet_glove_taskweave_mr2hd_activitynet.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/resnet'] -t_feat_dir: features/ActivityNet/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_mr2hd/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/activitynet/resnet_glove_tr_detr_activitynet.yml b/configs/activitynet/resnet_glove_tr_detr_activitynet.yml deleted file mode 100644 index 4f50906..0000000 --- a/configs/activitynet/resnet_glove_tr_detr_activitynet.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: tr_detr -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/resnet'] -t_feat_dir: features/ActivityNet/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_tr_detr/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/activitynet/resnet_glove_uvcom_activitynet.yml b/configs/activitynet/resnet_glove_uvcom_activitynet.yml deleted file mode 100644 index 406f8dc..0000000 --- a/configs/activitynet/resnet_glove_uvcom_activitynet.yml +++ /dev/null @@ -1,32 +0,0 @@ -model_name: uvcom -dset_name: activitynet -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/activitynet/activitynet_train_release.jsonl -eval_path: data/activitynet/activitynet_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/ActivityNet/resnet'] -t_feat_dir: features/ActivityNet/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_uvcom/activitynet -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/base.yml b/configs/base.yml index 3c9f04a..e90e928 100644 --- a/configs/base.yml +++ b/configs/base.yml @@ -38,5 +38,12 @@ label_loss_coef: 4 eos_coef: 0.1 lw_saliency: 1 -model_ema: False # If you use EMA, set True (Taskweave uses EMA) -ema_decay: 0.9 \ No newline at end of file +ckpt_filename: best.ckpt +train_log_filename: train.log +eval_log_filename: val.log +eval_split_name: val +aux_loss: True + +model_ema: False +ema_decay: 0.9 +results_dir: results \ No newline at end of file diff --git a/configs/charades/clip_cg_detr_charades.yml b/configs/charades/clip_cg_detr_charades.yml deleted file mode 100644 index ea77e17..0000000 --- a/configs/charades/clip_cg_detr_charades.yml +++ /dev/null @@ -1,37 +0,0 @@ -model_name: cg_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_cg_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/charades/clip_eatr_charades.yml b/configs/charades/clip_eatr_charades.yml deleted file mode 100644 index 11e9e87..0000000 --- a/configs/charades/clip_eatr_charades.yml +++ /dev/null @@ -1,25 +0,0 @@ -model_name: eatr -dset_name: charades -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_eatr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/charades/clip_moment_detr_charades.yml b/configs/charades/clip_moment_detr_charades.yml deleted file mode 100644 index bb69c1a..0000000 --- a/configs/charades/clip_moment_detr_charades.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: moment_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_moment_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/clip_qd_detr_charades.yml b/configs/charades/clip_qd_detr_charades.yml deleted file mode 100644 index ebd8327..0000000 --- a/configs/charades/clip_qd_detr_charades.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: qd_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_qd_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/clip_slowfast_cg_detr_charades.yml b/configs/charades/clip_slowfast_cg_detr_charades.yml deleted file mode 100644 index 64a7eb1..0000000 --- a/configs/charades/clip_slowfast_cg_detr_charades.yml +++ /dev/null @@ -1,37 +0,0 @@ -model_name: cg_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip', 'features/Charades/slowfast'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_cg_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/charades/clip_slowfast_eatr_charades.yml b/configs/charades/clip_slowfast_eatr_charades.yml deleted file mode 100644 index 8b22732..0000000 --- a/configs/charades/clip_slowfast_eatr_charades.yml +++ /dev/null @@ -1,25 +0,0 @@ -model_name: eatr -dset_name: charades -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip', 'features/Charades/slowfast'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_eatr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/charades/clip_slowfast_moment_detr_charades.yml b/configs/charades/clip_slowfast_moment_detr_charades.yml deleted file mode 100644 index a83d071..0000000 --- a/configs/charades/clip_slowfast_moment_detr_charades.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: moment_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip', 'features/Charades/slowfast'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_moment_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/clip_slowfast_qd_detr_charades.yml b/configs/charades/clip_slowfast_qd_detr_charades.yml deleted file mode 100644 index 8d0f1b1..0000000 --- a/configs/charades/clip_slowfast_qd_detr_charades.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: qd_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip', 'features/Charades/slowfast'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_qd_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/clip_slowfast_taskweave_hd2mr_charades.yml b/configs/charades/clip_slowfast_taskweave_hd2mr_charades.yml deleted file mode 100644 index 0cf5fc3..0000000 --- a/configs/charades/clip_slowfast_taskweave_hd2mr_charades.yml +++ /dev/null @@ -1,27 +0,0 @@ -model_name: taskweave -dset_name: charades -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip', 'features/Charades/slowfast'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_hd2mr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/clip_slowfast_taskweave_mr2hd_charades.yml b/configs/charades/clip_slowfast_taskweave_mr2hd_charades.yml deleted file mode 100644 index 9d70d0b..0000000 --- a/configs/charades/clip_slowfast_taskweave_mr2hd_charades.yml +++ /dev/null @@ -1,27 +0,0 @@ -model_name: taskweave -dset_name: charades -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip', 'features/Charades/slowfast'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_mr2hd/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/clip_slowfast_tr_detr_charades.yml b/configs/charades/clip_slowfast_tr_detr_charades.yml deleted file mode 100644 index 20c0643..0000000 --- a/configs/charades/clip_slowfast_tr_detr_charades.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: tr_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip', 'features/Charades/slowfast'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_tr_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/charades/clip_slowfast_uvcom_charades.yml b/configs/charades/clip_slowfast_uvcom_charades.yml deleted file mode 100644 index e74904c..0000000 --- a/configs/charades/clip_slowfast_uvcom_charades.yml +++ /dev/null @@ -1,36 +0,0 @@ -model_name: uvcom -dset_name: charades -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip', 'features/Charades/slowfast'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_uvcom/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/charades/clip_taskweave_hd2mr_charades.yml b/configs/charades/clip_taskweave_hd2mr_charades.yml deleted file mode 100644 index 56ad3da..0000000 --- a/configs/charades/clip_taskweave_hd2mr_charades.yml +++ /dev/null @@ -1,27 +0,0 @@ -model_name: taskweave -dset_name: charades -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_hd2mr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/clip_taskweave_mr2hd_charades.yml b/configs/charades/clip_taskweave_mr2hd_charades.yml deleted file mode 100644 index 1ab6ecf..0000000 --- a/configs/charades/clip_taskweave_mr2hd_charades.yml +++ /dev/null @@ -1,27 +0,0 @@ -model_name: taskweave -dset_name: charades -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_mr2hd/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/clip_tr_detr_charades.yml b/configs/charades/clip_tr_detr_charades.yml deleted file mode 100644 index 996f247..0000000 --- a/configs/charades/clip_tr_detr_charades.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: tr_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_tr_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/charades/clip_uvcom_charades.yml b/configs/charades/clip_uvcom_charades.yml deleted file mode 100644 index 3673e3b..0000000 --- a/configs/charades/clip_uvcom_charades.yml +++ /dev/null @@ -1,36 +0,0 @@ -model_name: uvcom -dset_name: charades -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/clip'] -t_feat_dir: features/Charades/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_uvcom/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/charades/resnet_glove_cg_detr_charades.yml b/configs/charades/resnet_glove_cg_detr_charades.yml deleted file mode 100644 index c432a7c..0000000 --- a/configs/charades/resnet_glove_cg_detr_charades.yml +++ /dev/null @@ -1,37 +0,0 @@ -model_name: cg_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/resnet'] -t_feat_dir: features/Charades/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_cg_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/charades/resnet_glove_eatr_charades.yml b/configs/charades/resnet_glove_eatr_charades.yml deleted file mode 100644 index 03b843d..0000000 --- a/configs/charades/resnet_glove_eatr_charades.yml +++ /dev/null @@ -1,25 +0,0 @@ -model_name: eatr -dset_name: charades -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/resnet'] -t_feat_dir: features/Charades/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_eatr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/charades/resnet_glove_moment_detr_charades.yml b/configs/charades/resnet_glove_moment_detr_charades.yml deleted file mode 100644 index de379d2..0000000 --- a/configs/charades/resnet_glove_moment_detr_charades.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: moment_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/resnet'] -t_feat_dir: features/Charades/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_moment_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/resnet_glove_qd_detr_charades.yml b/configs/charades/resnet_glove_qd_detr_charades.yml deleted file mode 100644 index 2e1f7f3..0000000 --- a/configs/charades/resnet_glove_qd_detr_charades.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: qd_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/resnet'] -t_feat_dir: features/Charades/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_qd_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/resnet_glove_taskweave_hd2mr_charades.yml b/configs/charades/resnet_glove_taskweave_hd2mr_charades.yml deleted file mode 100644 index c5d7f0c..0000000 --- a/configs/charades/resnet_glove_taskweave_hd2mr_charades.yml +++ /dev/null @@ -1,27 +0,0 @@ -model_name: taskweave -dset_name: charades -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/resnet'] -t_feat_dir: features/Charades/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_hd2mr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/resnet_glove_taskweave_mr2hd_charades.yml b/configs/charades/resnet_glove_taskweave_mr2hd_charades.yml deleted file mode 100644 index c3f6bb8..0000000 --- a/configs/charades/resnet_glove_taskweave_mr2hd_charades.yml +++ /dev/null @@ -1,27 +0,0 @@ -model_name: taskweave -dset_name: charades -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/resnet'] -t_feat_dir: features/Charades/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_mr2hd/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 \ No newline at end of file diff --git a/configs/charades/resnet_glove_tr_detr_charades.yml b/configs/charades/resnet_glove_tr_detr_charades.yml deleted file mode 100644 index 99b1304..0000000 --- a/configs/charades/resnet_glove_tr_detr_charades.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: tr_detr -dset_name: charades -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/resnet'] -t_feat_dir: features/Charades/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/clip_tr_detr/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/charades/resnet_glove_uvcom_charades.yml b/configs/charades/resnet_glove_uvcom_charades.yml deleted file mode 100644 index 5cde017..0000000 --- a/configs/charades/resnet_glove_uvcom_charades.yml +++ /dev/null @@ -1,36 +0,0 @@ -model_name: uvcom -dset_name: charades -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/charades/charades_train_release.jsonl -eval_path: data/charades/charades_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/Charades/resnet'] -t_feat_dir: features/Charades/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_uvcom/charades -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs_new/dataset/activitynet.yml b/configs/dataset/activitynet.yml similarity index 100% rename from configs_new/dataset/activitynet.yml rename to configs/dataset/activitynet.yml diff --git a/configs_new/dataset/charades.yml b/configs/dataset/charades.yml similarity index 100% rename from configs_new/dataset/charades.yml rename to configs/dataset/charades.yml diff --git a/configs_new/dataset/qvhighlight.yml b/configs/dataset/qvhighlight.yml similarity index 100% rename from configs_new/dataset/qvhighlight.yml rename to configs/dataset/qvhighlight.yml diff --git a/configs_new/dataset/tacos.yml b/configs/dataset/tacos.yml similarity index 100% rename from configs_new/dataset/tacos.yml rename to configs/dataset/tacos.yml diff --git a/configs_new/dataset/tvsum.yml b/configs/dataset/tvsum.yml similarity index 70% rename from configs_new/dataset/tvsum.yml rename to configs/dataset/tvsum.yml index 257650f..035d4b5 100644 --- a/configs_new/dataset/tvsum.yml +++ b/configs/dataset/tvsum.yml @@ -8,4 +8,15 @@ seed: 2018 lr: 0.001 lr_drop: 2000 n_epoch: 2000 -bsz: 4 \ No newline at end of file +bsz: 4 +domains: + - BK + - BT + - DS + - FM + - GA + - MS + - PK + - PR + - VT + - VU \ No newline at end of file diff --git a/configs_new/dataset/youtube_highlight.yml b/configs/dataset/youtube_highlight.yml similarity index 74% rename from configs_new/dataset/youtube_highlight.yml rename to configs/dataset/youtube_highlight.yml index 7aef189..a9abac8 100644 --- a/configs_new/dataset/youtube_highlight.yml +++ b/configs/dataset/youtube_highlight.yml @@ -9,4 +9,11 @@ seed: 2018 lr: 0.0002 lr_drop: 2000 n_epoch: 1000 -lw_saliency: 4 \ No newline at end of file +lw_saliency: 4 +domains: + - dog + - gymnastics + - parkour + - skating + - skiing + - surfing \ No newline at end of file diff --git a/configs_new/feature/clip.yml b/configs/feature/clip.yml similarity index 100% rename from configs_new/feature/clip.yml rename to configs/feature/clip.yml diff --git a/configs_new/feature/clip_slowfast.yml b/configs/feature/clip_slowfast.yml similarity index 100% rename from configs_new/feature/clip_slowfast.yml rename to configs/feature/clip_slowfast.yml diff --git a/configs_new/feature/clip_slowfast_pann.yml b/configs/feature/clip_slowfast_pann.yml similarity index 100% rename from configs_new/feature/clip_slowfast_pann.yml rename to configs/feature/clip_slowfast_pann.yml diff --git a/configs_new/feature/i3d_clip.yml b/configs/feature/i3d_clip.yml similarity index 100% rename from configs_new/feature/i3d_clip.yml rename to configs/feature/i3d_clip.yml diff --git a/configs_new/feature/resnet_glove.yml b/configs/feature/resnet_glove.yml similarity index 100% rename from configs_new/feature/resnet_glove.yml rename to configs/feature/resnet_glove.yml diff --git a/configs_new/model/cg_detr.yml b/configs/model/cg_detr.yml similarity index 100% rename from configs_new/model/cg_detr.yml rename to configs/model/cg_detr.yml diff --git a/configs_new/model/eatr.yml b/configs/model/eatr.yml similarity index 100% rename from configs_new/model/eatr.yml rename to configs/model/eatr.yml diff --git a/configs_new/model/moment_detr.yml b/configs/model/moment_detr.yml similarity index 100% rename from configs_new/model/moment_detr.yml rename to configs/model/moment_detr.yml diff --git a/configs_new/model/qd_detr.yml b/configs/model/qd_detr.yml similarity index 100% rename from configs_new/model/qd_detr.yml rename to configs/model/qd_detr.yml diff --git a/configs_new/model/taskweave_hd2mr.yml b/configs/model/taskweave_hd2mr.yml similarity index 100% rename from configs_new/model/taskweave_hd2mr.yml rename to configs/model/taskweave_hd2mr.yml diff --git a/configs_new/model/taskweave_mr2hd.yml b/configs/model/taskweave_mr2hd.yml similarity index 100% rename from configs_new/model/taskweave_mr2hd.yml rename to configs/model/taskweave_mr2hd.yml diff --git a/configs_new/model/tr_detr.yml b/configs/model/tr_detr.yml similarity index 100% rename from configs_new/model/tr_detr.yml rename to configs/model/tr_detr.yml diff --git a/configs_new/model/uvcom.yml b/configs/model/uvcom.yml similarity index 100% rename from configs_new/model/uvcom.yml rename to configs/model/uvcom.yml diff --git a/configs/qvhighlight/clip_cg_detr_qvhighlight.yml b/configs/qvhighlight/clip_cg_detr_qvhighlight.yml deleted file mode 100644 index 653ebe3..0000000 --- a/configs/qvhighlight/clip_cg_detr_qvhighlight.yml +++ /dev/null @@ -1,34 +0,0 @@ -model_name: cg_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_cg_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/qvhighlight/clip_eatr_qvhighlight.yml b/configs/qvhighlight/clip_eatr_qvhighlight.yml deleted file mode 100644 index cc52573..0000000 --- a/configs/qvhighlight/clip_eatr_qvhighlight.yml +++ /dev/null @@ -1,21 +0,0 @@ -model_name: eatr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_eatr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/qvhighlight/clip_moment_detr_qvhighlight.yml b/configs/qvhighlight/clip_moment_detr_qvhighlight.yml deleted file mode 100644 index 72d557c..0000000 --- a/configs/qvhighlight/clip_moment_detr_qvhighlight.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: moment_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_moment_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/qvhighlight/clip_qd_detr_qvhighlight.yml b/configs/qvhighlight/clip_qd_detr_qvhighlight.yml deleted file mode 100644 index 541d01b..0000000 --- a/configs/qvhighlight/clip_qd_detr_qvhighlight.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: qd_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_qd_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_cg_detr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_cg_detr_qvhighlight.yml deleted file mode 100644 index 8269ed1..0000000 --- a/configs/qvhighlight/clip_slowfast_cg_detr_qvhighlight.yml +++ /dev/null @@ -1,34 +0,0 @@ -model_name: cg_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_cg_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_eatr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_eatr_qvhighlight.yml deleted file mode 100644 index f966542..0000000 --- a/configs/qvhighlight/clip_slowfast_eatr_qvhighlight.yml +++ /dev/null @@ -1,21 +0,0 @@ -model_name: eatr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_eatr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_moment_detr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_moment_detr_qvhighlight.yml deleted file mode 100644 index d94501d..0000000 --- a/configs/qvhighlight/clip_slowfast_moment_detr_qvhighlight.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: moment_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_moment_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_moment_detr_qvhighlight_pretrain.yml b/configs/qvhighlight/clip_slowfast_moment_detr_qvhighlight_pretrain.yml deleted file mode 100644 index 0e9d359..0000000 --- a/configs/qvhighlight/clip_slowfast_moment_detr_qvhighlight_pretrain.yml +++ /dev/null @@ -1,19 +0,0 @@ -model_name: moment_detr -dset_name: qvhighlight_pretrain -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/qvhighlight/subs_train.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -t_feat_dir: features/QVHighlight/clip_text_subs_train -t_feat_dir_eval: features/QVHighlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_moment_detr_pretrain/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_pann_cg_detr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_pann_cg_detr_qvhighlight.yml deleted file mode 100644 index c82017d..0000000 --- a/configs/qvhighlight/clip_slowfast_pann_cg_detr_qvhighlight.yml +++ /dev/null @@ -1,37 +0,0 @@ -model_name: cg_detr -dset_name: qvhighlight -ctx_mode: video_audio_tef -v_feat_types: slowfast_clip -a_feat_types: pann -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -a_feat_dirs: ['features/QVHighlight/pann'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -a_feat_dim: 2048 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_pann_cg_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_pann_eatr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_pann_eatr_qvhighlight.yml deleted file mode 100644 index 99b5317..0000000 --- a/configs/qvhighlight/clip_slowfast_pann_eatr_qvhighlight.yml +++ /dev/null @@ -1,24 +0,0 @@ -model_name: eatr -dset_name: qvhighlight -ctx_mode: video_audio_tef -v_feat_types: slowfast_clip -a_feat_types: pann -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -a_feat_dirs: ['features/QVHighlight/pann'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -a_feat_dim: 2048 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_pann_eatr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_pann_moment_detr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_pann_moment_detr_qvhighlight.yml deleted file mode 100644 index d3356a7..0000000 --- a/configs/qvhighlight/clip_slowfast_pann_moment_detr_qvhighlight.yml +++ /dev/null @@ -1,21 +0,0 @@ -model_name: moment_detr -dset_name: qvhighlight -ctx_mode: video_audio_tef -v_feat_types: slowfast_clip -a_feat_types: pann -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -a_feat_dirs: ['features/QVHighlight/pann'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -a_feat_dim: 2048 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_pann_moment_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_pann_qd_detr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_pann_qd_detr_qvhighlight.yml deleted file mode 100644 index 3af8734..0000000 --- a/configs/qvhighlight/clip_slowfast_pann_qd_detr_qvhighlight.yml +++ /dev/null @@ -1,21 +0,0 @@ -model_name: qd_detr -dset_name: qvhighlight -ctx_mode: video_audio_tef -v_feat_types: slowfast_clip -a_feat_types: pann -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -a_feat_dirs: ['features/QVHighlight/pann'] -t_feat_dir: 'features/QVHighlight/clip_text' -v_feat_dim: 2818 -a_feat_dim: 2048 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_pann_qd_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_pann_taskweave_hd2mr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_pann_taskweave_hd2mr_qvhighlight.yml deleted file mode 100644 index 6d015c6..0000000 --- a/configs/qvhighlight/clip_slowfast_pann_taskweave_hd2mr_qvhighlight.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: taskweave -dset_name: qvhighlight -ctx_mode: video_audio_tef -v_feat_types: slowfast_clip -a_feat_types: pann -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -a_feat_dirs: ['features/QVHighlight/pann'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -a_feat_dim: 2048 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_pann_taskweave_hd2mr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_pann_taskweave_mr2hd_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_pann_taskweave_mr2hd_qvhighlight.yml deleted file mode 100644 index d7aa9f4..0000000 --- a/configs/qvhighlight/clip_slowfast_pann_taskweave_mr2hd_qvhighlight.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: taskweave -dset_name: qvhighlight -ctx_mode: video_audio_tef -v_feat_types: slowfast_clip -a_feat_types: pann -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -a_feat_dirs: ['features/QVHighlight/pann'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -a_feat_dim: 2048 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_pann_taskweave_mr2hd/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_pann_tr_detr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_pann_tr_detr_qvhighlight.yml deleted file mode 100644 index 41f29f0..0000000 --- a/configs/qvhighlight/clip_slowfast_pann_tr_detr_qvhighlight.yml +++ /dev/null @@ -1,25 +0,0 @@ -model_name: tr_detr -dset_name: qvhighlight -ctx_mode: video_audio_tef -v_feat_types: slowfast_clip -a_feat_types: pann -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -a_feat_dirs: ['features/QVHighlight/pann'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -a_feat_dim: 2048 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_pann_tr_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_pann_uvcom_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_pann_uvcom_qvhighlight.yml deleted file mode 100644 index 4da8c93..0000000 --- a/configs/qvhighlight/clip_slowfast_pann_uvcom_qvhighlight.yml +++ /dev/null @@ -1,35 +0,0 @@ -model_name: uvcom -dset_name: qvhighlight -ctx_mode: video_audio_tef -v_feat_types: slowfast_clip -a_feat_types: pann -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -a_feat_dirs: ['features/QVHighlight/pann'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -a_feat_dim: 2048 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_pann_uvcom/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_qd_detr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_qd_detr_qvhighlight.yml deleted file mode 100644 index e7597ca..0000000 --- a/configs/qvhighlight/clip_slowfast_qd_detr_qvhighlight.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: qd_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_qd_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_taskweave_hd2mr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_taskweave_hd2mr_qvhighlight.yml deleted file mode 100644 index 220987c..0000000 --- a/configs/qvhighlight/clip_slowfast_taskweave_hd2mr_qvhighlight.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_hd2mr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_taskweave_mr2hd_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_taskweave_mr2hd_qvhighlight.yml deleted file mode 100644 index 2c00b3e..0000000 --- a/configs/qvhighlight/clip_slowfast_taskweave_mr2hd_qvhighlight.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_mr2hd/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_tr_detr_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_tr_detr_qvhighlight.yml deleted file mode 100644 index 93e58c1..0000000 --- a/configs/qvhighlight/clip_slowfast_tr_detr_qvhighlight.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: tr_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_tr_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/qvhighlight/clip_slowfast_uvcom_qvhighlight.yml b/configs/qvhighlight/clip_slowfast_uvcom_qvhighlight.yml deleted file mode 100644 index 61a0543..0000000 --- a/configs/qvhighlight/clip_slowfast_uvcom_qvhighlight.yml +++ /dev/null @@ -1,32 +0,0 @@ -model_name: uvcom -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip', 'features/QVHighlight/slowfast'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_uvcom/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/qvhighlight/clip_taskweave_hd2mr_qvhighlight.yml b/configs/qvhighlight/clip_taskweave_hd2mr_qvhighlight.yml deleted file mode 100644 index 9c23179..0000000 --- a/configs/qvhighlight/clip_taskweave_hd2mr_qvhighlight.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_hd2mr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/qvhighlight/clip_taskweave_mr2hd_qvhighlight.yml b/configs/qvhighlight/clip_taskweave_mr2hd_qvhighlight.yml deleted file mode 100644 index 5892598..0000000 --- a/configs/qvhighlight/clip_taskweave_mr2hd_qvhighlight.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_mr2hd/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/qvhighlight/clip_tr_detr_qvhighlight.yml b/configs/qvhighlight/clip_tr_detr_qvhighlight.yml deleted file mode 100644 index 4f04dde..0000000 --- a/configs/qvhighlight/clip_tr_detr_qvhighlight.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: tr_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_tr_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/qvhighlight/clip_uvcom_qvhighlight.yml b/configs/qvhighlight/clip_uvcom_qvhighlight.yml deleted file mode 100644 index ab10d20..0000000 --- a/configs/qvhighlight/clip_uvcom_qvhighlight.yml +++ /dev/null @@ -1,32 +0,0 @@ -model_name: uvcom -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/clip'] -t_feat_dir: features/QVHighlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_uvcom/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/qvhighlight/resnet_glove_cg_detr_qvhighlight.yml b/configs/qvhighlight/resnet_glove_cg_detr_qvhighlight.yml deleted file mode 100644 index 229473f..0000000 --- a/configs/qvhighlight/resnet_glove_cg_detr_qvhighlight.yml +++ /dev/null @@ -1,34 +0,0 @@ -model_name: cg_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/resnet'] -t_feat_dir: features/QVHighlight/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_cg_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 45 -total_prompts: 10 -num_prompts: 2 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/qvhighlight/resnet_glove_eatr_qvhighlight.yml b/configs/qvhighlight/resnet_glove_eatr_qvhighlight.yml deleted file mode 100644 index dd2bf97..0000000 --- a/configs/qvhighlight/resnet_glove_eatr_qvhighlight.yml +++ /dev/null @@ -1,21 +0,0 @@ -model_name: eatr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/resnet'] -t_feat_dir: features/QVHighlight/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_slowfast_eatr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/qvhighlight/resnet_glove_moment_detr_qvhighlight.yml b/configs/qvhighlight/resnet_glove_moment_detr_qvhighlight.yml deleted file mode 100644 index 9d4bd59..0000000 --- a/configs/qvhighlight/resnet_glove_moment_detr_qvhighlight.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: moment_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/resnet'] -t_feat_dir: features/QVHighlight/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_moment_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/qvhighlight/resnet_glove_qd_detr_qvhighlight.yml b/configs/qvhighlight/resnet_glove_qd_detr_qvhighlight.yml deleted file mode 100644 index e7fc41b..0000000 --- a/configs/qvhighlight/resnet_glove_qd_detr_qvhighlight.yml +++ /dev/null @@ -1,18 +0,0 @@ -model_name: qd_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/resnet'] -t_feat_dir: features/QVHighlight/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_qd_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 \ No newline at end of file diff --git a/configs/qvhighlight/resnet_glove_taskweave_hd2mr_qvhighlight.yml b/configs/qvhighlight/resnet_glove_taskweave_hd2mr_qvhighlight.yml deleted file mode 100644 index ab1b8d8..0000000 --- a/configs/qvhighlight/resnet_glove_taskweave_hd2mr_qvhighlight.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/resnet'] -t_feat_dir: features/QVHighlight/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_hd2mr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/qvhighlight/resnet_glove_taskweave_mr2hd_qvhighlight.yml b/configs/qvhighlight/resnet_glove_taskweave_mr2hd_qvhighlight.yml deleted file mode 100644 index ab1c321..0000000 --- a/configs/qvhighlight/resnet_glove_taskweave_mr2hd_qvhighlight.yml +++ /dev/null @@ -1,23 +0,0 @@ -model_name: taskweave -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/resnet'] -t_feat_dir: features/QVHighlight/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_mr2hd/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/qvhighlight/resnet_glove_tr_detr_qvhighlight.yml b/configs/qvhighlight/resnet_glove_tr_detr_qvhighlight.yml deleted file mode 100644 index 1ce7fd6..0000000 --- a/configs/qvhighlight/resnet_glove_tr_detr_qvhighlight.yml +++ /dev/null @@ -1,22 +0,0 @@ -model_name: tr_detr -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/resnet'] -t_feat_dir: features/QVHighlight/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_tr_detr/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/qvhighlight/resnet_glove_uvcom_qvhighlight.yml b/configs/qvhighlight/resnet_glove_uvcom_qvhighlight.yml deleted file mode 100644 index 1140995..0000000 --- a/configs/qvhighlight/resnet_glove_uvcom_qvhighlight.yml +++ /dev/null @@ -1,32 +0,0 @@ -model_name: uvcom -dset_name: qvhighlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/qvhighlight/highlight_train_release.jsonl -eval_path: data/qvhighlight/highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/QVHighlight/resnet'] -t_feat_dir: features/QVHighlight/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_uvcom/qvhighlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/tacos/clip_cg_detr_tacos.yml b/configs/tacos/clip_cg_detr_tacos.yml deleted file mode 100644 index b764820..0000000 --- a/configs/tacos/clip_cg_detr_tacos.yml +++ /dev/null @@ -1,41 +0,0 @@ -model_name: cg_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_cg_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 200 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 50 -total_prompts: 10 -num_prompts: 2 - -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 - -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_eatr_tacos.yml b/configs/tacos/clip_eatr_tacos.yml deleted file mode 100644 index df497dd..0000000 --- a/configs/tacos/clip_eatr_tacos.yml +++ /dev/null @@ -1,29 +0,0 @@ -model_name: eatr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_eatr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/tacos/clip_moment_detr_tacos.yml b/configs/tacos/clip_moment_detr_tacos.yml deleted file mode 100644 index c64ab25..0000000 --- a/configs/tacos/clip_moment_detr_tacos.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: moment_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_moment_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_qd_detr_tacos.yml b/configs/tacos/clip_qd_detr_tacos.yml deleted file mode 100644 index ee1428c..0000000 --- a/configs/tacos/clip_qd_detr_tacos.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: qd_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_qd_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_slowfast_cg_detr_tacos.yml b/configs/tacos/clip_slowfast_cg_detr_tacos.yml deleted file mode 100644 index 7527a46..0000000 --- a/configs/tacos/clip_slowfast_cg_detr_tacos.yml +++ /dev/null @@ -1,41 +0,0 @@ -model_name: cg_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip', 'features/tacos/slowfast'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_cg_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 200 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 50 -total_prompts: 10 -num_prompts: 2 - -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 - -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_slowfast_eatr_tacos.yml b/configs/tacos/clip_slowfast_eatr_tacos.yml deleted file mode 100644 index dd1a491..0000000 --- a/configs/tacos/clip_slowfast_eatr_tacos.yml +++ /dev/null @@ -1,29 +0,0 @@ -model_name: eatr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip', 'features/tacos/slowfast'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_eatr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/tacos/clip_slowfast_moment_detr_tacos.yml b/configs/tacos/clip_slowfast_moment_detr_tacos.yml deleted file mode 100644 index 33f8006..0000000 --- a/configs/tacos/clip_slowfast_moment_detr_tacos.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: moment_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip', 'features/tacos/slowfast'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_moment_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_slowfast_qd_detr_tacos.yml b/configs/tacos/clip_slowfast_qd_detr_tacos.yml deleted file mode 100644 index d99dc40..0000000 --- a/configs/tacos/clip_slowfast_qd_detr_tacos.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: qd_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip', 'features/tacos/slowfast'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_qd_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_slowfast_taskweave_hd2mr_tacos.yml b/configs/tacos/clip_slowfast_taskweave_hd2mr_tacos.yml deleted file mode 100644 index 48d3834..0000000 --- a/configs/tacos/clip_slowfast_taskweave_hd2mr_tacos.yml +++ /dev/null @@ -1,31 +0,0 @@ -model_name: taskweave -dset_name: tacos -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip', 'features/tacos/slowfast'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_hd2mr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_slowfast_taskweave_mr2hd_tacos.yml b/configs/tacos/clip_slowfast_taskweave_mr2hd_tacos.yml deleted file mode 100644 index 884d29a..0000000 --- a/configs/tacos/clip_slowfast_taskweave_mr2hd_tacos.yml +++ /dev/null @@ -1,31 +0,0 @@ -model_name: taskweave -dset_name: tacos -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip', 'features/tacos/slowfast'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_mr2hd/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_slowfast_tr_detr_tacos.yml b/configs/tacos/clip_slowfast_tr_detr_tacos.yml deleted file mode 100644 index ac65526..0000000 --- a/configs/tacos/clip_slowfast_tr_detr_tacos.yml +++ /dev/null @@ -1,30 +0,0 @@ -model_name: tr_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip', 'features/tacos/slowfast'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_tr_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/tacos/clip_slowfast_uvcom_tacos.yml b/configs/tacos/clip_slowfast_uvcom_tacos.yml deleted file mode 100644 index befe512..0000000 --- a/configs/tacos/clip_slowfast_uvcom_tacos.yml +++ /dev/null @@ -1,40 +0,0 @@ -model_name: qd_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip', 'features/tacos/slowfast'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_uvcom/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/tacos/clip_taskweave_hd2mr_tacos.yml b/configs/tacos/clip_taskweave_hd2mr_tacos.yml deleted file mode 100644 index 66699d9..0000000 --- a/configs/tacos/clip_taskweave_hd2mr_tacos.yml +++ /dev/null @@ -1,31 +0,0 @@ -model_name: taskweave -dset_name: tacos -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_hd2mr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_taskweave_mr2hd_tacos.yml b/configs/tacos/clip_taskweave_mr2hd_tacos.yml deleted file mode 100644 index d6ded3b..0000000 --- a/configs/tacos/clip_taskweave_mr2hd_tacos.yml +++ /dev/null @@ -1,31 +0,0 @@ -model_name: taskweave -dset_name: tacos -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_mr2hd/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/clip_tr_detr_tacos.yml b/configs/tacos/clip_tr_detr_tacos.yml deleted file mode 100644 index b986b8c..0000000 --- a/configs/tacos/clip_tr_detr_tacos.yml +++ /dev/null @@ -1,30 +0,0 @@ -model_name: qd_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_tr_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/tacos/clip_uvcom_tacos.yml b/configs/tacos/clip_uvcom_tacos.yml deleted file mode 100644 index 7a20aeb..0000000 --- a/configs/tacos/clip_uvcom_tacos.yml +++ /dev/null @@ -1,40 +0,0 @@ -model_name: qd_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/clip'] -t_feat_dir: features/tacos/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_uvcom/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/tacos/resnet_glove_cg_detr_tacos.yml b/configs/tacos/resnet_glove_cg_detr_tacos.yml deleted file mode 100644 index e25cc61..0000000 --- a/configs/tacos/resnet_glove_cg_detr_tacos.yml +++ /dev/null @@ -1,41 +0,0 @@ -model_name: cg_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/resnet'] -t_feat_dir: features/tacos/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_cg_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 200 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 -num_dummies: 50 -total_prompts: 10 -num_prompts: 2 - -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 - -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/resnet_glove_eatr_tacos.yml b/configs/tacos/resnet_glove_eatr_tacos.yml deleted file mode 100644 index 1e3cf29..0000000 --- a/configs/tacos/resnet_glove_eatr_tacos.yml +++ /dev/null @@ -1,29 +0,0 @@ -model_name: eatr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/resnet'] -t_feat_dir: features/tacos/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_eatr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# charades-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/tacos/resnet_glove_moment_detr_tacos.yml b/configs/tacos/resnet_glove_moment_detr_tacos.yml deleted file mode 100644 index ed97256..0000000 --- a/configs/tacos/resnet_glove_moment_detr_tacos.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: moment_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/resnet'] -t_feat_dir: features/tacos/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_moment_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/resnet_glove_qd_detr_tacos.yml b/configs/tacos/resnet_glove_qd_detr_tacos.yml deleted file mode 100644 index ea3783f..0000000 --- a/configs/tacos/resnet_glove_qd_detr_tacos.yml +++ /dev/null @@ -1,26 +0,0 @@ -model_name: qd_detr -dset_name: tacos -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/resnet'] -t_feat_dir: features/tacos/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_qd_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/resnet_glove_taskweave_hd2mr_tacos.yml b/configs/tacos/resnet_glove_taskweave_hd2mr_tacos.yml deleted file mode 100644 index bf39c3f..0000000 --- a/configs/tacos/resnet_glove_taskweave_hd2mr_tacos.yml +++ /dev/null @@ -1,31 +0,0 @@ -model_name: taskweave -dset_name: tacos -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/resnet'] -t_feat_dir: features/tacos/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_hd2mr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/resnet_glove_taskweave_mr2hd_tacos.yml b/configs/tacos/resnet_glove_taskweave_mr2hd_tacos.yml deleted file mode 100644 index ebb5460..0000000 --- a/configs/tacos/resnet_glove_taskweave_mr2hd_tacos.yml +++ /dev/null @@ -1,31 +0,0 @@ -model_name: taskweave -dset_name: tacos -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/resnet'] -t_feat_dir: features/tacos/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_mr2hd/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 \ No newline at end of file diff --git a/configs/tacos/resnet_glove_tr_detr_tacos.yml b/configs/tacos/resnet_glove_tr_detr_tacos.yml deleted file mode 100644 index 6045ed1..0000000 --- a/configs/tacos/resnet_glove_tr_detr_tacos.yml +++ /dev/null @@ -1,30 +0,0 @@ -model_name: taskweave -dset_name: tacos -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/resnet'] -t_feat_dir: features/tacos/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_tr_detr/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 - -# TR-DETR specific losses -VTC_loss_coef: 0.3 -CTC_loss_coef: 0.5 \ No newline at end of file diff --git a/configs/tacos/resnet_glove_uvcom_tacos.yml b/configs/tacos/resnet_glove_uvcom_tacos.yml deleted file mode 100644 index 69e6c1e..0000000 --- a/configs/tacos/resnet_glove_uvcom_tacos.yml +++ /dev/null @@ -1,40 +0,0 @@ -model_name: taskweave -dset_name: tacos -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tacos/tacos_train_release.jsonl -eval_path: data/tacos/tacos_test_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tacos/resnet'] -t_feat_dir: features/tacos/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_uvcom/tacos -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 - -# tacos-specific params -lw_saliency: 4 -lr: 0.0002 -lr_drop: 40 -max_v_l: -1 -max_q_l: 100 -eval_bsz: 32 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -lr_drop: 100 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/tvsum/clip_cgdetr_tvsum.yml b/configs/tvsum/clip_cgdetr_tvsum.yml deleted file mode 100644 index 9eec0ee..0000000 --- a/configs/tvsum/clip_cgdetr_tvsum.yml +++ /dev/null @@ -1,43 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: cg_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_cg_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 - -num_dummies: 3 -total_prompts: 10 -num_prompts: 1 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/tvsum/clip_eatr_tvsum.yml b/configs/tvsum/clip_eatr_tvsum.yml deleted file mode 100644 index 2513a2e..0000000 --- a/configs/tvsum/clip_eatr_tvsum.yml +++ /dev/null @@ -1,29 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: eatr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_eatr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/tvsum/clip_moment_detr_tvsum.yml b/configs/tvsum/clip_moment_detr_tvsum.yml deleted file mode 100644 index 503d3e6..0000000 --- a/configs/tvsum/clip_moment_detr_tvsum.yml +++ /dev/null @@ -1,26 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: moment_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_moment_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 \ No newline at end of file diff --git a/configs/tvsum/clip_qd_detr_tvsum.yml b/configs/tvsum/clip_qd_detr_tvsum.yml deleted file mode 100644 index 92674b2..0000000 --- a/configs/tvsum/clip_qd_detr_tvsum.yml +++ /dev/null @@ -1,26 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: qd_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_qd_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 \ No newline at end of file diff --git a/configs/tvsum/clip_slowfast_cgdetr_tvsum.yml b/configs/tvsum/clip_slowfast_cgdetr_tvsum.yml deleted file mode 100644 index 7b7f96e..0000000 --- a/configs/tvsum/clip_slowfast_cgdetr_tvsum.yml +++ /dev/null @@ -1,43 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: cg_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip', 'features/tvsum/slowfast'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_cg_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 - -num_dummies: 3 -total_prompts: 10 -num_prompts: 1 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/tvsum/clip_slowfast_eatr_tvsum.yml b/configs/tvsum/clip_slowfast_eatr_tvsum.yml deleted file mode 100644 index d438c86..0000000 --- a/configs/tvsum/clip_slowfast_eatr_tvsum.yml +++ /dev/null @@ -1,29 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: eatr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip', 'features/tvsum/slowfast'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_eatr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/tvsum/clip_slowfast_moment_detr_tvsum.yml b/configs/tvsum/clip_slowfast_moment_detr_tvsum.yml deleted file mode 100644 index 415a464..0000000 --- a/configs/tvsum/clip_slowfast_moment_detr_tvsum.yml +++ /dev/null @@ -1,26 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: moment_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip', 'features/tvsum/slowfast'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_moment_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 \ No newline at end of file diff --git a/configs/tvsum/clip_slowfast_qd_detr_tvsum.yml b/configs/tvsum/clip_slowfast_qd_detr_tvsum.yml deleted file mode 100644 index c0f63c6..0000000 --- a/configs/tvsum/clip_slowfast_qd_detr_tvsum.yml +++ /dev/null @@ -1,26 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: qd_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip', 'features/tvsum/slowfast'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_qd_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 \ No newline at end of file diff --git a/configs/tvsum/clip_slowfast_taskweave_hd2mr_tvsum.yml b/configs/tvsum/clip_slowfast_taskweave_hd2mr_tvsum.yml deleted file mode 100644 index 6ed9cb5..0000000 --- a/configs/tvsum/clip_slowfast_taskweave_hd2mr_tvsum.yml +++ /dev/null @@ -1,31 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip', 'features/tvsum/slowfast'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_hd2mr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/tvsum/clip_slowfast_taskweave_mr2hd_tvsum.yml b/configs/tvsum/clip_slowfast_taskweave_mr2hd_tvsum.yml deleted file mode 100644 index 7545602..0000000 --- a/configs/tvsum/clip_slowfast_taskweave_mr2hd_tvsum.yml +++ /dev/null @@ -1,31 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip', 'features/tvsum/slowfast'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_mr2hd/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/tvsum/clip_slowfast_uvcom_tvsum.yml b/configs/tvsum/clip_slowfast_uvcom_tvsum.yml deleted file mode 100644 index 4cde22f..0000000 --- a/configs/tvsum/clip_slowfast_uvcom_tvsum.yml +++ /dev/null @@ -1,39 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: uvcom -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip', 'features/tvsum/slowfast'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_uvcom/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/tvsum/clip_taskweave_hd2mr_tvsum.yml b/configs/tvsum/clip_taskweave_hd2mr_tvsum.yml deleted file mode 100644 index 5c2dabf..0000000 --- a/configs/tvsum/clip_taskweave_hd2mr_tvsum.yml +++ /dev/null @@ -1,31 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_hd2mr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/tvsum/clip_taskweave_mr2hd_tvsum.yml b/configs/tvsum/clip_taskweave_mr2hd_tvsum.yml deleted file mode 100644 index 441d44c..0000000 --- a/configs/tvsum/clip_taskweave_mr2hd_tvsum.yml +++ /dev/null @@ -1,31 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_mr2hd/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/tvsum/clip_uvcom_tvsum.yml b/configs/tvsum/clip_uvcom_tvsum.yml deleted file mode 100644 index 15e6458..0000000 --- a/configs/tvsum/clip_uvcom_tvsum.yml +++ /dev/null @@ -1,39 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: uvcom -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/clip'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_uvcom/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/tvsum/i3d_clip_cgdetr_tvsum.yml b/configs/tvsum/i3d_clip_cgdetr_tvsum.yml deleted file mode 100644 index 27f62c5..0000000 --- a/configs/tvsum/i3d_clip_cgdetr_tvsum.yml +++ /dev/null @@ -1,43 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: cg_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: i3d -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/i3d'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2050 -t_feat_dim: 512 -aux_loss: True -results_dir: results/i3d_clip_cg_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 - -num_dummies: 3 -total_prompts: 10 -num_prompts: 1 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/tvsum/i3d_clip_eatr_tvsum.yml b/configs/tvsum/i3d_clip_eatr_tvsum.yml deleted file mode 100644 index 9a04ef9..0000000 --- a/configs/tvsum/i3d_clip_eatr_tvsum.yml +++ /dev/null @@ -1,29 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: eatr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: i3d -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/i3d'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2050 -t_feat_dim: 512 -aux_loss: True -results_dir: results/i3d_clip_eatr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/tvsum/i3d_clip_moment_detr_tvsum.yml b/configs/tvsum/i3d_clip_moment_detr_tvsum.yml deleted file mode 100644 index 95dcef1..0000000 --- a/configs/tvsum/i3d_clip_moment_detr_tvsum.yml +++ /dev/null @@ -1,26 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: moment_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: i3d -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/i3d'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2050 -t_feat_dim: 512 -aux_loss: True -results_dir: results/i3d_clip_moment_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 \ No newline at end of file diff --git a/configs/tvsum/i3d_clip_qd_detr_tvsum.yml b/configs/tvsum/i3d_clip_qd_detr_tvsum.yml deleted file mode 100644 index ff76efc..0000000 --- a/configs/tvsum/i3d_clip_qd_detr_tvsum.yml +++ /dev/null @@ -1,26 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: qd_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: i3d -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/i3d'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2050 -t_feat_dim: 512 -aux_loss: True -results_dir: results/i3d_clip_qd_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 \ No newline at end of file diff --git a/configs/tvsum/i3d_clip_taskweave_hd2mr_tvsum.yml b/configs/tvsum/i3d_clip_taskweave_hd2mr_tvsum.yml deleted file mode 100644 index cc1fe24..0000000 --- a/configs/tvsum/i3d_clip_taskweave_hd2mr_tvsum.yml +++ /dev/null @@ -1,31 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: i3d -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/i3d'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2050 -t_feat_dim: 512 -aux_loss: True -results_dir: results/i3d_clip_taskweave_hd2mr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/tvsum/i3d_clip_taskweave_mr2hd_tvsum.yml b/configs/tvsum/i3d_clip_taskweave_mr2hd_tvsum.yml deleted file mode 100644 index b1b51c9..0000000 --- a/configs/tvsum/i3d_clip_taskweave_mr2hd_tvsum.yml +++ /dev/null @@ -1,31 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: i3d -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/i3d'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2050 -t_feat_dim: 512 -aux_loss: True -results_dir: results/i3d_clip_taskweave_mr2hd/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/tvsum/i3d_clip_tr_detr_tvsum.yml b/configs/tvsum/i3d_clip_tr_detr_tvsum.yml deleted file mode 100644 index 92d4457..0000000 --- a/configs/tvsum/i3d_clip_tr_detr_tvsum.yml +++ /dev/null @@ -1,30 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: tr_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: i3d -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/i3d'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2050 -t_feat_dim: 512 -aux_loss: True -results_dir: results/i3d_clip_tr_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# TR-DETR specific losses -VTC_loss_coef: 0.5 -CTC_loss_coef: 2 \ No newline at end of file diff --git a/configs/tvsum/i3d_clip_uvcom_tvsum.yml b/configs/tvsum/i3d_clip_uvcom_tvsum.yml deleted file mode 100644 index cc5fffe..0000000 --- a/configs/tvsum/i3d_clip_uvcom_tvsum.yml +++ /dev/null @@ -1,39 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: uvcom -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: i3d -t_feat_type: clip -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/i3d'] -t_feat_dir: features/tvsum/clip_text -v_feat_dim: 2050 -t_feat_dim: 512 -aux_loss: True -results_dir: results/i3d_clip_uvcom/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/tvsum/resnet_glove_cgdetr_tvsum.yml b/configs/tvsum/resnet_glove_cgdetr_tvsum.yml deleted file mode 100644 index 2664020..0000000 --- a/configs/tvsum/resnet_glove_cgdetr_tvsum.yml +++ /dev/null @@ -1,43 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: cg_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/resnet'] -t_feat_dir: features/tvsum/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_cg_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 - -num_dummies: 3 -total_prompts: 10 -num_prompts: 1 - -lw_saliency: 1 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/tvsum/resnet_glove_eatr_tvsum.yml b/configs/tvsum/resnet_glove_eatr_tvsum.yml deleted file mode 100644 index 86d8800..0000000 --- a/configs/tvsum/resnet_glove_eatr_tvsum.yml +++ /dev/null @@ -1,29 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: eatr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/resnet'] -t_feat_dir: features/tvsum/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_eatr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/tvsum/resnet_glove_moment_detr_tvsum.yml b/configs/tvsum/resnet_glove_moment_detr_tvsum.yml deleted file mode 100644 index d8566c9..0000000 --- a/configs/tvsum/resnet_glove_moment_detr_tvsum.yml +++ /dev/null @@ -1,26 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: moment_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/resnet'] -t_feat_dir: features/tvsum/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_moment_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 \ No newline at end of file diff --git a/configs/tvsum/resnet_glove_qd_detr_tvsum.yml b/configs/tvsum/resnet_glove_qd_detr_tvsum.yml deleted file mode 100644 index 36fd4e8..0000000 --- a/configs/tvsum/resnet_glove_qd_detr_tvsum.yml +++ /dev/null @@ -1,26 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: qd_detr -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/resnet'] -t_feat_dir: features/tvsum/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_qd_detr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 \ No newline at end of file diff --git a/configs/tvsum/resnet_glove_taskweave_hd2mr_tvsum.yml b/configs/tvsum/resnet_glove_taskweave_hd2mr_tvsum.yml deleted file mode 100644 index fefb771..0000000 --- a/configs/tvsum/resnet_glove_taskweave_hd2mr_tvsum.yml +++ /dev/null @@ -1,31 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/resnet'] -t_feat_dir: features/tvsum/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_hd2mr/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/tvsum/resnet_glove_taskweave_mr2hd_tvsum.yml b/configs/tvsum/resnet_glove_taskweave_mr2hd_tvsum.yml deleted file mode 100644 index 5b71555..0000000 --- a/configs/tvsum/resnet_glove_taskweave_mr2hd_tvsum.yml +++ /dev/null @@ -1,31 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/resnet'] -t_feat_dir: features/tvsum/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_taskweave_mr2hd/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/tvsum/resnet_glove_uvcom_tvsum.yml b/configs/tvsum/resnet_glove_uvcom_tvsum.yml deleted file mode 100644 index 2819b74..0000000 --- a/configs/tvsum/resnet_glove_uvcom_tvsum.yml +++ /dev/null @@ -1,39 +0,0 @@ -seed: 2018 # CGDETR TVSum -model_name: uvcom -dset_name: tvsum -ctx_mode: video_tef -v_feat_types: resnet -t_feat_type: glove -train_path: data/tvsum/tvsum_train_release.jsonl -eval_path: data/tvsum/tvsum_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/tvsum/resnet'] -t_feat_dir: features/tvsum/glove -v_feat_dim: 2050 -t_feat_dim: 300 -aux_loss: True -results_dir: results/resnet_glove_uvcom/tvsum -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 2 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.001 -lr_drop: 2000 -n_epoch: 2000 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/youtube_highlight/clip_cgdetr_youtube_highlight.yml b/configs/youtube_highlight/clip_cgdetr_youtube_highlight.yml deleted file mode 100644 index c5216a4..0000000 --- a/configs/youtube_highlight/clip_cgdetr_youtube_highlight.yml +++ /dev/null @@ -1,43 +0,0 @@ -seed: 2018 -model_name: cg_detr -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_cg_detr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.0002 -lr_drop: 2000 -n_epoch: 1000 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 - -num_dummies: 1 -total_prompts: 10 -num_prompts: 1 - -lw_saliency: 4 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/youtube_highlight/clip_eatr_youtube_highlight.yml b/configs/youtube_highlight/clip_eatr_youtube_highlight.yml deleted file mode 100644 index ffd4369..0000000 --- a/configs/youtube_highlight/clip_eatr_youtube_highlight.yml +++ /dev/null @@ -1,30 +0,0 @@ -seed: 2018 -model_name: eatr -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_eatr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 2000 -n_epoch: 1000 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/youtube_highlight/clip_moment_detr_youtube_highlight.yml b/configs/youtube_highlight/clip_moment_detr_youtube_highlight.yml deleted file mode 100644 index 8009d2d..0000000 --- a/configs/youtube_highlight/clip_moment_detr_youtube_highlight.yml +++ /dev/null @@ -1,27 +0,0 @@ -seed: 2018 -model_name: moment_detr -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_moment_detr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 2000 -n_epoch: 1000 \ No newline at end of file diff --git a/configs/youtube_highlight/clip_qd_detr_youtube_highlight.yml b/configs/youtube_highlight/clip_qd_detr_youtube_highlight.yml deleted file mode 100644 index 9f3f365..0000000 --- a/configs/youtube_highlight/clip_qd_detr_youtube_highlight.yml +++ /dev/null @@ -1,27 +0,0 @@ -seed: 2018 -model_name: qd_detr -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_qd_detr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.001 -lr_drop: 2000 -n_epoch: 1000 \ No newline at end of file diff --git a/configs/youtube_highlight/clip_slowfast_cgdetr_youtube_highlight.yml b/configs/youtube_highlight/clip_slowfast_cgdetr_youtube_highlight.yml deleted file mode 100644 index 09f8592..0000000 --- a/configs/youtube_highlight/clip_slowfast_cgdetr_youtube_highlight.yml +++ /dev/null @@ -1,43 +0,0 @@ -seed: 2018 -model_name: cg_detr -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip', 'features/youtube_highlight/slowfast'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_cg_detr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lr: 0.0002 -lr_drop: 2000 -n_epoch: 1000 - -# CG-DETR-specific parameters -enc_layers: 3 -dec_layers: 3 -t2v_layers: 2 -sent_layers: 1 -moment_layers: 1 -dummy_layers: 2 - -num_dummies: 1 -total_prompts: 10 -num_prompts: 1 - -lw_saliency: 4 -lw_wattn: 1 -lw_ms_align: 1 -lw_distill: 1 \ No newline at end of file diff --git a/configs/youtube_highlight/clip_slowfast_eatr_youtube_highlight.yml b/configs/youtube_highlight/clip_slowfast_eatr_youtube_highlight.yml deleted file mode 100644 index 11c6a32..0000000 --- a/configs/youtube_highlight/clip_slowfast_eatr_youtube_highlight.yml +++ /dev/null @@ -1,30 +0,0 @@ -seed: 2018 -model_name: eatr -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip', 'features/youtube_highlight/slowfast'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_eatr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 2000 -n_epoch: 1000 - -# EaTR-specific params -event_coef: 3 \ No newline at end of file diff --git a/configs/youtube_highlight/clip_slowfast_moment_detr_youtube_highlight.yml b/configs/youtube_highlight/clip_slowfast_moment_detr_youtube_highlight.yml deleted file mode 100644 index f350d22..0000000 --- a/configs/youtube_highlight/clip_slowfast_moment_detr_youtube_highlight.yml +++ /dev/null @@ -1,27 +0,0 @@ -seed: 2018 -model_name: moment_detr -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip', 'features/youtube_highlight/slowfast'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_moment_detr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 2000 -n_epoch: 1000 \ No newline at end of file diff --git a/configs/youtube_highlight/clip_slowfast_qd_detr_youtube_highlight.yml b/configs/youtube_highlight/clip_slowfast_qd_detr_youtube_highlight.yml deleted file mode 100644 index 3157d7e..0000000 --- a/configs/youtube_highlight/clip_slowfast_qd_detr_youtube_highlight.yml +++ /dev/null @@ -1,27 +0,0 @@ -seed: 2018 -model_name: qd_detr -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip', 'features/youtube_highlight/slowfast'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_qd_detr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 2000 -n_epoch: 1000 \ No newline at end of file diff --git a/configs/youtube_highlight/clip_slowfast_taskweave_hd2mr_youtube_highlight.yml b/configs/youtube_highlight/clip_slowfast_taskweave_hd2mr_youtube_highlight.yml deleted file mode 100644 index 9f1f66a..0000000 --- a/configs/youtube_highlight/clip_slowfast_taskweave_hd2mr_youtube_highlight.yml +++ /dev/null @@ -1,32 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip', 'features/youtube_highlight/slowfast'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_hd2mr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.001 -lr_drop: 2000 -n_epoch: 1000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/youtube_highlight/clip_slowfast_taskweave_mr2hd_youtube_highlight.yml b/configs/youtube_highlight/clip_slowfast_taskweave_mr2hd_youtube_highlight.yml deleted file mode 100644 index 85b6698..0000000 --- a/configs/youtube_highlight/clip_slowfast_taskweave_mr2hd_youtube_highlight.yml +++ /dev/null @@ -1,32 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip', 'features/youtube_highlight/slowfast'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_taskweave_mr2hd/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.001 -lr_drop: 2000 -n_epoch: 1000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/youtube_highlight/clip_slowfast_uvcom_youtube_highlight.yml b/configs/youtube_highlight/clip_slowfast_uvcom_youtube_highlight.yml deleted file mode 100644 index 1644067..0000000 --- a/configs/youtube_highlight/clip_slowfast_uvcom_youtube_highlight.yml +++ /dev/null @@ -1,40 +0,0 @@ -seed: 2018 -model_name: uvcom -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: slowfast_clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip', 'features/youtube_highlight/slowfast'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 2818 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_slowfast_uvcom/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 2000 -n_epoch: 1000 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs/youtube_highlight/clip_taskweave_hd2mr_youtube_highlight.yml b/configs/youtube_highlight/clip_taskweave_hd2mr_youtube_highlight.yml deleted file mode 100644 index d48d3f3..0000000 --- a/configs/youtube_highlight/clip_taskweave_hd2mr_youtube_highlight.yml +++ /dev/null @@ -1,32 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_hd2mr/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.001 -lr_drop: 2000 -n_epoch: 1000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: False -model_ema: True \ No newline at end of file diff --git a/configs/youtube_highlight/clip_taskweave_mr2hd_youtube_highlight.yml b/configs/youtube_highlight/clip_taskweave_mr2hd_youtube_highlight.yml deleted file mode 100644 index 9668fc7..0000000 --- a/configs/youtube_highlight/clip_taskweave_mr2hd_youtube_highlight.yml +++ /dev/null @@ -1,32 +0,0 @@ -seed: 2018 -model_name: taskweave -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_taskweave_mr2hd/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.001 -lr_drop: 2000 -n_epoch: 1000 - -# Taskweave-specific; This model needs to train MR2HD and HD2MR for MR and HD, respectively. -# If mr2hd is False, the strategy is HD2MR -mr2hd: True -model_ema: True \ No newline at end of file diff --git a/configs/youtube_highlight/clip_uvcom_youtube_highlight.yml b/configs/youtube_highlight/clip_uvcom_youtube_highlight.yml deleted file mode 100644 index 9ad56fd..0000000 --- a/configs/youtube_highlight/clip_uvcom_youtube_highlight.yml +++ /dev/null @@ -1,40 +0,0 @@ -seed: 2018 -model_name: uvcom -dset_name: youtube_highlight -ctx_mode: video_tef -v_feat_types: clip -t_feat_type: clip -train_path: data/youtube_highlight/youtube_highlight_train_release.jsonl -eval_path: data/youtube_highlight/youtube_highlight_val_release.jsonl -eval_split_name: val -v_feat_dirs: ['features/youtube_highlight/clip'] -t_feat_dir: features/youtube_highlight/clip_text -v_feat_dim: 514 -t_feat_dim: 512 -aux_loss: True -results_dir: results/clip_uvcom/youtube_highlight -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -clip_length: 1 -bsz: 4 -max_q_l: 100 -max_v_l: 1000 - -lw_saliency: 4 -lr: 0.0002 -lr_drop: 2000 -n_epoch: 1000 - -# uvcom-specific params -enc_layers: 3 -dec_layers: 3 -num_queries: 30 -n_txt_mu: 5 -n_visual_mu: 30 -sim_loss_coef: 1 -neg_loss_coef: 0.5 -lr_gamma: 0.1 -neg_choose_epoch: 80 -em_iter: 5 -cross_fusion: False \ No newline at end of file diff --git a/configs_new/base.yml b/configs_new/base.yml deleted file mode 100644 index 425d2a2..0000000 --- a/configs_new/base.yml +++ /dev/null @@ -1,48 +0,0 @@ -seed: 2023 -device: 'cuda' -num_workers: 4 - -lr: 0.0001 -lr_drop: 400 -wd: 0.0001 -n_epoch: 200 -max_es_cnt: 200 -bsz: 32 -eval_bsz: 100 -grad_clip: 0.1 -max_q_l: 32 -max_v_l: 75 -max_windows: 5 -clip_length: 1 -eval_epoch_interval: 1 - -position_embedding: "sine" -enc_layers: 2 -dec_layers: 2 -dim_feedforward: 1024 -hidden_dim: 256 -input_dropout: 0.5 -dropout: 0.1 -nheads: 8 -num_queries: 10 - -n_input_proj: 2 -saliency_margin: 0.2 -span_loss_type: "l1" -set_cost_span: 10 -set_cost_giou: 1 -set_cost_class: 4 -span_loss_coef: 10 -giou_loss_coef: 1 -label_loss_coef: 4 -eos_coef: 0.1 -lw_saliency: 1 - -ckpt_filename: best.ckpt -train_log_filename: train.log -eval_log_filename: val.log -eval_split_name: val -aux_loss: True - -model_ema: False -ema_decay: 0.9 \ No newline at end of file diff --git a/training/config.py b/training/config.py index 5e9edf7..8ecee94 100755 --- a/training/config.py +++ b/training/config.py @@ -18,40 +18,84 @@ import time import torch import argparse - -from lighthouse.common.utils.basic_utils import mkdirp, load_json, save_json, make_zipfile, dict_to_markdown import shutil import yaml +from lighthouse.common.utils.basic_utils import mkdirp, load_json, save_json, make_zipfile, dict_to_markdown from easydict import EasyDict class BaseOptions(object): - def __init__(self): - pass + def __init__(self, model, dataset, feature): + self.model = model + self.dataset = dataset + self.feature = feature + self.opt = {} - def parse(self, yaml_path, domain): - opt = {} + @property + def option(self): + if len(self.opt) == 0: + raise RuntimeError('option is empty. Did you run parse()?') + return self.opt - # base yaml - with open('configs/base.yml', 'r') as f: - yml = yaml.load(f, Loader=yaml.FullLoader) - opt.update(yml) - - with open('{}'.format(yaml_path), 'r') as f: + def update(self, yaml_file): + with open(yaml_file, 'r') as f: yml = yaml.load(f, Loader=yaml.FullLoader) - opt.update(yml) - - opt = EasyDict(opt) - opt.domain = domain - if opt.domain: - opt.results_dir = os.path.join(opt.results_dir, opt.domain) - opt.ckpt_filepath = os.path.join(opt.results_dir, opt.ckpt_filename) - opt.train_log_filepath = os.path.join(opt.results_dir, opt.train_log_filename) - opt.eval_log_filepath = os.path.join(opt.results_dir, opt.eval_log_filename) - os.makedirs(opt.results_dir, exist_ok=True) - else: - opt.ckpt_filepath = os.path.join(opt.results_dir, opt.ckpt_filename) - opt.train_log_filepath = os.path.join(opt.results_dir, opt.train_log_filename) - opt.eval_log_filepath = os.path.join(opt.results_dir, opt.eval_log_filename) - os.makedirs(opt.results_dir, exist_ok=True) - return opt \ No newline at end of file + self.opt.update(yml) + + def parse(self): + base_cfg = 'configs/base.yml' + feature_cfg = f'configs/feature/{self.feature}.yml' + model_cfg = f'configs/model/{self.model}.yml' + dataset_cfg = f'configs/dataset/{self.dataset}.yml' + cfgs = [base_cfg, feature_cfg, model_cfg, dataset_cfg] + for cfg in cfgs: + self.update(cfg) + + self.opt = EasyDict(self.opt) + + # result directory + self.opt.results_dir = os.path.join(self.opt.results_dir, self.model, self.dataset, self.feature) + self.opt.ckpt_filepath = os.path.join(self.opt.results_dir, self.opt.ckpt_filename) + self.opt.train_log_filepath = os.path.join(self.opt.results_dir, self.opt.train_log_filename) + self.opt.eval_log_filepath = os.path.join(self.opt.results_dir, self.opt.eval_log_filename) + + # feature directory + v_feat_dirs = None + t_feat_dir = None + a_feat_dirs = None + a_feat_types = None + + if self.feature == 'clip_slowfast_pann': + v_feat_dirs = [f'features/{self.dataset}/clip', f'features/{self.dataset}/slowfast'] + t_feat_dir = f'features/{self.dataset}/clip_text' + a_feat_dirs = [f'features/{self.dataset}/pann'] + a_feat_types = self.opt.a_feat_types + + elif self.feature == 'clip_slowfast': + v_feat_dirs = [f'features/{self.dataset}/clip', f'features/{self.dataset}/slowfast'] + t_feat_dir = f'features/{self.dataset}/clip_text' + + elif self.feature == 'clip': + v_feat_dirs = [f'features/{self.dataset}/clip'] + t_feat_dir = f'features/{self.dataset}/clip_text' + + elif self.feature == 'resnet_glove': + v_feat_dirs = [f'features/{self.dataset}/resnet'] + t_feat_dir = f'features/{self.dataset}/glove' + + elif self.feature == 'i3d_clip': + v_feat_dirs = [f'features/{self.dataset}/i3d'] + t_feat_dir = f'features/{self.dataset}/clip_text' + + self.opt.v_feat_dirs = v_feat_dirs + self.opt.t_feat_dir = t_feat_dir + self.opt.a_feat_dirs = a_feat_dirs + self.opt.a_feat_types = a_feat_types + + def makedirs(self): + if 'results_dir' not in self.opt: + raise RuntimeError('results_dir is not set in self.opt. Did you run parse()?') + os.makedirs(self.opt.results_dir, exist_ok=True) + if 'domains' in self.opt: + for domain in self.domains: + os.makedirs(os.path.join(self.opt.results_dir, domain), exist_ok=True) \ No newline at end of file diff --git a/training/__init__.py b/training/standalone_eval/__init__.py similarity index 100% rename from training/__init__.py rename to training/standalone_eval/__init__.py diff --git a/training/train.py b/training/train.py index d81af54..cf11d8f 100755 --- a/training/train.py +++ b/training/train.py @@ -52,10 +52,13 @@ from easydict import EasyDict -from training.config import BaseOptions -from training.dataset import StartEndDataset, start_end_collate, prepare_batch_inputs -from training.cg_detr_dataset import CGDETR_StartEndDataset, cg_detr_start_end_collate, cg_detr_prepare_batch_inputs -from training.evaluate import eval_epoch, start_inference, setup_model +import sys +sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))) + +from config import BaseOptions +from dataset import StartEndDataset, start_end_collate, prepare_batch_inputs +from cg_detr_dataset import CGDETR_StartEndDataset, cg_detr_start_end_collate, cg_detr_prepare_batch_inputs +from evaluate import eval_epoch, start_inference, setup_model from lighthouse.common.utils.basic_utils import AverageMeter, dict_to_markdown, write_log, save_checkpoint, rename_latest_to_best from lighthouse.common.utils.model_utils import count_parameters, ModelEMA @@ -206,23 +209,22 @@ def train(model, criterion, optimizer, lr_scheduler, train_dataset, val_dataset, rename_latest_to_best(latest_file_paths) -def main(yaml_path, pretrained_model_path, domain): +def main(opt, resume=None, domain=None): logger.info("Setup config, data and model...") - opt = BaseOptions().parse(yaml_path, domain) set_seed(opt.seed) # dataset & data loader dataset_config = EasyDict( dset_name=opt.dset_name, - domain=opt.domain, + domain=domain, data_path=opt.train_path, ctx_mode=opt.ctx_mode, v_feat_dirs=opt.v_feat_dirs, - a_feat_dirs=opt.a_feat_dirs if "a_feat_dirs" in opt else [], + a_feat_dirs=opt.a_feat_dirs, q_feat_dir=opt.t_feat_dir, q_feat_type="last_hidden_state", v_feat_types=opt.v_feat_types, - a_feat_types=opt.a_feat_types if "a_feat_types" in opt else None, + a_feat_types=opt.a_feat_types, max_q_l=opt.max_q_l, max_v_l=opt.max_v_l, clip_len=opt.clip_length, @@ -240,11 +242,13 @@ def main(yaml_path, pretrained_model_path, domain): # prepare model model, criterion, optimizer, lr_scheduler = setup_model(opt) logger.info(f"Model {model}") - # load checkpoint - if pretrained_model_path is not None: - checkpoint = torch.load(pretrained_model_path) + + # load checkpoint for QVHighlight pretrain -> finetune + if resume is not None: + checkpoint = torch.load(resume) model.load_state_dict(checkpoint["model"]) - logger.info("Model checkpoint: {}".format(pretrained_model_path)) + logger.info("Loaded model checkpoint: {}".format(resume)) + count_parameters(model) logger.info("Start Training...") @@ -252,13 +256,49 @@ def main(yaml_path, pretrained_model_path, domain): train(model, criterion, optimizer, lr_scheduler, train_dataset, eval_dataset, opt) +def check_valid_combination(dataset, feature): + if feature == 'i3d_clip': + return dataset == 'tvsum' + + if feature == 'clip_slowfast_pann': + return dataset == 'qvhighlight' or dataset == 'qvhighlight_pretrain' + + if dataset == 'youtube_highlight': + # Due to unavailable access to the original videos, we publish only CLIP and CLIP+Slowfast for YouTube Highlight. + return dataset != 'resnet_glove' + + return True + + if __name__ == '__main__': parser = argparse.ArgumentParser() - parser.add_argument('--config', type=str, required=True, help='yaml config path for training. e.g., configs/qd_detr_qvhighlight.yml') - parser.add_argument('--pretrained_model_path', type=str, help='saved model path', default=None) - parser.add_argument('--domain', type=str, help='training domain for TVSum and YouTube Highlights . e.g., BK and dog. Note that they are not necessary for other datasets') + parser.add_argument('--model', '-m', type=str, required=True, + choices=['moment_detr', 'qd_detr', 'eatr', 'cg_detr', 'uvcom', 'tr_detr', 'taskweave_hd2mr', 'taskweave_mr2hd'], + help='model name. select from [moment_detr, qd_detr, eatr, cg_detr, uvcom, tr_detr, taskweave_hd2mr, taskweave_mr2hd]') + parser.add_argument('--dataset', '-d', type=str, required=True, + choices=['activitynet', 'charades', 'qvhighlight', 'qvhighlight_pretrain', 'tacos', 'tvsum', 'youtube_highlight'], + help='dataset name. select from [activitynet, charades, qvhighlight, qvhighlight_pretrain, tacos, tvsum, youtube_highlight]') + parser.add_argument('--feature', '-f', type=str, required=True, + choices=['resnet_glove', 'clip', 'clip_slowfast', 'clip_slowfast_pann', 'i3d_clip'], + help='feature name. select from [resnet_glove, clip, clip_slowfast, clip_slowfast_pann, i3d_clip].' + 'NOTE: i3d_clip and clip_slowfast_pann are only for TVSum and QVHighlight, respectively') + parser.add_argument('--resume', '-r', type=str, help='specify model path for fine-tuning. If None, train the model from scratch.') args = parser.parse_args() - yaml_path = args.config - pretrained_model_path = args.pretrained_model_path - domain = args.domain - main(yaml_path, pretrained_model_path, domain) \ No newline at end of file + + is_valid = check_valid_combination(args.dataset, args.feature) + + if is_valid: + option_manager = BaseOptions(args.model, args.dataset, args.feature) + option_manager.parse() + option_manager.makedirs() + opt = option_manager.option + + if 'domains' in opt: + for domain in opt.domains: + self.opt.results_dir = os.path.join(self.opt.results_dir, domain) + main(opt, resume=args.resume, domain=domain) + else: + main(opt, resume=args.resume) + + else: + raise ValueError('The combination of dataset and feature is invalid: dataset={}, feature={}'.format(args.dataset, args.feature)) \ No newline at end of file From 633c70d385e0acbfaad9f19013270423f428d949 Mon Sep 17 00:00:00 2001 From: awkrail Date: Thu, 5 Sep 2024 16:39:44 +0900 Subject: [PATCH 5/8] fix readme --- README.md | 49 +++++++++++++++++++++++++++---------------------- 1 file changed, 27 insertions(+), 22 deletions(-) diff --git a/README.md b/README.md index 54db947..30d67c8 100755 --- a/README.md +++ b/README.md @@ -182,40 +182,45 @@ lighthouse/ ### Training and evaluation #### Training -The general training command is: +The training command is: ``` -PYTHONPATH=. python training/train.py --config configs/DATASET/FEATURE_MODEL_DATASET.yml +python training/train.py --model MODEL --dataset DATASET --feature FEATURE [--resume RESUME] ``` -| | Options | -|---------|--------------------------------------------------------------------| -| Model | moment_detr, qd_detr, eatr, cg_detr, uvcom, tr_detr, taskweave | -| Feature | resnet_glove, clip, clip_slowfast, clip_slowfast_pann | -| Dataset | qvhighlight, activitynet, charades, tacos, tvsum, youtube_highlight| +| | Options | +|---------|------------------------------------------------------------------------------------------| +| Model | moment_detr, qd_detr, eatr, cg_detr, uvcom, tr_detr, taskweave_mr2hd, taskweave_hd2mr | +| Feature | resnet_glove, clip, clip_slowfast, clip_slowfast_pann, i3d_clip | +| Dataset | qvhighlight, qvhighlight_pretrain, activitynet, charades, tacos, tvsum, youtube_highlight| -For example, to train moment_detr on QVHighlights with CLIP+Slowfast features, run: +(**Example 1**) Moment DETR w/ CLIP+Slowfast on QVHighlights: ``` -PYTHONPATH=. python training/train.py --config configs/qvhighlight/clip_slowfast_moment_detr_qvhighlight.yml +python training/train.py --model moment_detr --dataset qvhighlight --feature clip_slowfast ``` -To train the models on HD datasets (i.e., TVSum and YouTube Highlight), you need to specify the domain.
-For example, to train moment_detr in BK domain on TVSum, run: +(**Example 2**) Moment DETR w/ CLIP+Slowfast+PANNs (Audio) on QVHighlights: ``` -PYTHONPATH=. python training/train.py --config configs/tvsum/clip_slowfast_moment_detr_tvsum.yml --domain BK +python training/train.py --model moment_detr --dataset qvhighlight --feature clip_slowfast_pann +``` +(**Pre-train & Fine-tuning**) Lighthouse supports pre-training. Run: +``` +python training/train.py --model moment_detr --dataset qvhighlight_pretrain --feature clip_slowfast +``` +Then fine-tune the model with `--resume` option: +``` +python training/train.py --model moment_detr --dataset qvhighlight --feature clip_slowfast --resume results/moment_detr/qvhighlight_pretrain/clip_slowfast/best.ckpt ``` #### Evaluation -The evaluation command is (in this example, we evaluate QD-DETR/CLIP+Slowfast on the QVHighlight val set): +The evaluation command is: +``` +python training/evaluate.py --model MODEL --dataset DATASET --feature FEATURE --split {val,test} --model_path MODEL_PATH +``` +(**Example 1**) Evaluating Moment DETR w/ CLIP+Slowfast on the QVHighlights val set: ``` -PYTHONPATH=. python training/evaluate.py --config configs/qvhighlight/clip_slowfast_qd_detr_qvhighlight.yml \ - --model_path results/clip_slowfast_qd_detr/qvhighlight/best.ckpt \ - --eval_split_name val \ - --eval_path data/qvhighlight/highlight_val_release.jsonl +python training/train.py --model moment_detr --dataset qvhighlight --feature clip_slowfast --split val --model_path results/moment_detr/qvhighlight/clip_slowfast/best.ckpt ``` -To generate submission files for QVHighlight test sets, run (**QVHighlights only**): +To generate submission files for QVHighlight test sets, change split into test (**QVHighlights only**): ``` -PYTHONPATH=. python training/evaluate.py --config configs/qvhighlight/clip_slowfast_qd_detr_qvhighlight.yml \ - --model_path results/clip_slowfast_qd_detr/qvhighlight/best.ckpt \ - --eval_split_name test \ - --eval_path data/qvhighlight/highlight_test_release.jsonl +python training/train.py --model moment_detr --dataset qvhighlight --feature clip_slowfast --split test --model_path results/moment_detr/qvhighlight/clip_slowfast/best.ckpt ``` Then zip `hl_val_submission.jsonl` and `hl_test_submission.jsonl`, and submit it to the [Codalab](https://codalab.lisn.upsaclay.fr/competitions/6937) (**QVHighlights only**): ``` From e306c4052c89654770765650b7a6a702ee4fd7e3 Mon Sep 17 00:00:00 2001 From: awkrail Date: Thu, 5 Sep 2024 16:41:33 +0900 Subject: [PATCH 6/8] fix readme --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 30d67c8..8e2ff21 100755 --- a/README.md +++ b/README.md @@ -200,7 +200,7 @@ python training/train.py --model moment_detr --dataset qvhighlight --feature cli ``` python training/train.py --model moment_detr --dataset qvhighlight --feature clip_slowfast_pann ``` -(**Pre-train & Fine-tuning**) Lighthouse supports pre-training. Run: +(**Pre-train & Fine-tuning, QVHighlights only**) Lighthouse supports pre-training. Run: ``` python training/train.py --model moment_detr --dataset qvhighlight_pretrain --feature clip_slowfast ``` From a5a333c319c7fa991f99e610479b4210bbb9b7fc Mon Sep 17 00:00:00 2001 From: awkrail Date: Thu, 5 Sep 2024 17:39:18 +0900 Subject: [PATCH 7/8] add qvhighlight pretrain --- configs/dataset/qvhighlight_pretrain.yml | 4 ++ training/config.py | 77 +++++++++++++++++------- training/train.py | 7 +-- 3 files changed, 61 insertions(+), 27 deletions(-) create mode 100644 configs/dataset/qvhighlight_pretrain.yml diff --git a/configs/dataset/qvhighlight_pretrain.yml b/configs/dataset/qvhighlight_pretrain.yml new file mode 100644 index 0000000..87ad87f --- /dev/null +++ b/configs/dataset/qvhighlight_pretrain.yml @@ -0,0 +1,4 @@ +dset_name: qvhighlight_pretrain +clip_length: 2 +train_path: data/qvhighlight/subs_train.jsonl +eval_path: data/qvhighlight/highlight_val_release.jsonl \ No newline at end of file diff --git a/training/config.py b/training/config.py index 8ecee94..35626df 100755 --- a/training/config.py +++ b/training/config.py @@ -64,38 +64,69 @@ def parse(self): t_feat_dir = None a_feat_dirs = None a_feat_types = None + t_feat_dir_pretrain_eval = None - if self.feature == 'clip_slowfast_pann': - v_feat_dirs = [f'features/{self.dataset}/clip', f'features/{self.dataset}/slowfast'] - t_feat_dir = f'features/{self.dataset}/clip_text' - a_feat_dirs = [f'features/{self.dataset}/pann'] - a_feat_types = self.opt.a_feat_types + if self.dataset == 'qvhighlight_pretrain': - elif self.feature == 'clip_slowfast': - v_feat_dirs = [f'features/{self.dataset}/clip', f'features/{self.dataset}/slowfast'] - t_feat_dir = f'features/{self.dataset}/clip_text' - - elif self.feature == 'clip': - v_feat_dirs = [f'features/{self.dataset}/clip'] - t_feat_dir = f'features/{self.dataset}/clip_text' - - elif self.feature == 'resnet_glove': - v_feat_dirs = [f'features/{self.dataset}/resnet'] - t_feat_dir = f'features/{self.dataset}/glove' - - elif self.feature == 'i3d_clip': - v_feat_dirs = [f'features/{self.dataset}/i3d'] - t_feat_dir = f'features/{self.dataset}/clip_text' + dataset = self.dataset.replace('_pretrain', '') + + if self.feature == 'clip_slowfast_pann': + v_feat_dirs = [f'features/{dataset}/clip', f'features/{dataset}/slowfast'] + t_feat_dir = f'features/{dataset}/clip_text_subs_train' + t_feat_dir_pretrain_eval = f'features/{dataset}/clip_text' + a_feat_dirs = [f'features/{dataset}/pann'] + a_feat_types = self.opt.a_feat_types + + elif self.feature == 'clip_slowfast': + v_feat_dirs = [f'features/{dataset}/clip', f'features/{dataset}/slowfast'] + t_feat_dir = f'features/{dataset}/clip_text_subs_train' + t_feat_dir_pretrain_eval = f'features/{dataset}/clip_text' + + elif self.feature == 'clip': + v_feat_dirs = [f'features/{dataset}/clip'] + t_feat_dir = f'features/{dataset}/clip_text_subs_train' + t_feat_dir_pretrain_eval = f'features/{dataset}/clip_text' + + else: + raise ValueError(f'For pre-train, features should include CLIP, but {self.feature} is used.') + + else: + if self.feature == 'clip_slowfast_pann': + v_feat_dirs = [f'features/{self.dataset}/clip', f'features/{self.dataset}/slowfast'] + t_feat_dir = f'features/{self.dataset}/clip_text' + a_feat_dirs = [f'features/{self.dataset}/pann'] + a_feat_types = self.opt.a_feat_types + + elif self.feature == 'clip_slowfast': + v_feat_dirs = [f'features/{self.dataset}/clip', f'features/{self.dataset}/slowfast'] + t_feat_dir = f'features/{self.dataset}/clip_text' + + elif self.feature == 'clip': + v_feat_dirs = [f'features/{self.dataset}/clip'] + t_feat_dir = f'features/{self.dataset}/clip_text' + + elif self.feature == 'resnet_glove': + v_feat_dirs = [f'features/{self.dataset}/resnet'] + t_feat_dir = f'features/{self.dataset}/glove' + + elif self.feature == 'i3d_clip': + v_feat_dirs = [f'features/{self.dataset}/i3d'] + t_feat_dir = f'features/{self.dataset}/clip_text' self.opt.v_feat_dirs = v_feat_dirs self.opt.t_feat_dir = t_feat_dir self.opt.a_feat_dirs = a_feat_dirs self.opt.a_feat_types = a_feat_types - - def makedirs(self): + self.opt.t_feat_dir_pretrain_eval = t_feat_dir_pretrain_eval + + def clean_and_makedirs(self): if 'results_dir' not in self.opt: raise RuntimeError('results_dir is not set in self.opt. Did you run parse()?') + + if os.path.exists(self.opt.results_dir): + shutil.rmtree(self.opt.results_dir) + os.makedirs(self.opt.results_dir, exist_ok=True) if 'domains' in self.opt: - for domain in self.domains: + for domain in self.opt.domains: os.makedirs(os.path.join(self.opt.results_dir, domain), exist_ok=True) \ No newline at end of file diff --git a/training/train.py b/training/train.py index cf11d8f..4c54108 100755 --- a/training/train.py +++ b/training/train.py @@ -109,7 +109,6 @@ def calculate_taskweave_losses(loss_dict, weight_dict, hd_log_var, mr_log_var): grouped_losses["loss_hd"].append(loss_dict[k]) loss_mr = sum(grouped_losses["loss_mr"]) loss_hd = sum(grouped_losses["loss_hd"]) - # hd_log_var, mr_log_var = hd_log_var.to(loss_hd.device), mr_log_var.to(loss_mr.device) losses = 2 * loss_hd * torch.exp(-hd_log_var) + 1 * loss_mr * torch.exp(-mr_log_var) + hd_log_var + mr_log_var return losses @@ -236,7 +235,7 @@ def main(opt, resume=None, domain=None): train_dataset = CGDETR_StartEndDataset(**dataset_config) if opt.model_name == 'cg_detr' else StartEndDataset(**dataset_config) copied_eval_config = copy.deepcopy(dataset_config) copied_eval_config.data_path = opt.eval_path - copied_eval_config.q_feat_dir = opt.t_feat_dir_eval if "t_feat_dir_eval" in opt else opt.t_feat_dir + copied_eval_config.q_feat_dir = opt.t_feat_dir_pretrain_eval if opt.t_feat_dir_pretrain_eval is not None else opt.t_feat_dir eval_dataset = CGDETR_StartEndDataset(**copied_eval_config) if opt.model_name == 'cg_detr' else StartEndDataset(**copied_eval_config) # prepare model @@ -290,12 +289,12 @@ def check_valid_combination(dataset, feature): if is_valid: option_manager = BaseOptions(args.model, args.dataset, args.feature) option_manager.parse() - option_manager.makedirs() + option_manager.clean_and_makedirs() opt = option_manager.option if 'domains' in opt: for domain in opt.domains: - self.opt.results_dir = os.path.join(self.opt.results_dir, domain) + opt.results_dir = os.path.join(opt.results_dir, domain) main(opt, resume=args.resume, domain=domain) else: main(opt, resume=args.resume) From a25816c5ae018ea6d638bb394dd1ad35e78d2ebd Mon Sep 17 00:00:00 2001 From: awkrail Date: Tue, 10 Sep 2024 18:00:28 +0900 Subject: [PATCH 8/8] fix clip_slowfast_pann qvh pretrain --- training/cg_detr_dataset.py | 11 ++++------- training/dataset.py | 13 +++++-------- training/postprocessing.py | 4 ---- 3 files changed, 9 insertions(+), 19 deletions(-) diff --git a/training/cg_detr_dataset.py b/training/cg_detr_dataset.py index 8fe8d20..b1bfa12 100644 --- a/training/cg_detr_dataset.py +++ b/training/cg_detr_dataset.py @@ -449,19 +449,16 @@ def _get_video_feat_by_vid(self, vid): def _get_audio_feat_by_vid(self, vid): a_feat_list = [] for _feat_dir in self.a_feat_dirs: - if self.dset_name == 'qvhighlight': - if self.a_feat_types == "clap": - _feat_path = join(_feat_dir, f"{vid}.npz") - _feat = np.load(_feat_path)["features"][:self.max_a_l].astype(np.float32) - elif self.a_feat_types == "pann": + if self.dset_name == 'qvhighlight' or self.dset_name == 'qvhighlight_pretrain': + if self.a_feat_types == "pann": _feat_path = join(_feat_dir, f"{vid}.npy") _feat = np.load(_feat_path)[:self.max_a_l].astype(np.float32) else: - raise NotImplementedError() + raise NotImplementedError _feat = l2_normalize_np_array(_feat) # normalize? a_feat_list.append(_feat) else: - raise NotImplementedError() + raise NotImplementedError # some features are slightly longer than the others min_len = min([len(e) for e in a_feat_list]) diff --git a/training/dataset.py b/training/dataset.py index c267592..7e3533e 100755 --- a/training/dataset.py +++ b/training/dataset.py @@ -215,7 +215,7 @@ def __getitem__(self, index): model_inputs["saliency_pos_labels"], model_inputs["saliency_neg_labels"], model_inputs["saliency_all_labels"] = \ self.get_saliency_labels_sub_as_query(meta["relevant_windows"][0], ctx_l) else: - raise NotImplementedError() + raise NotImplementedError return dict(meta=meta, model_inputs=model_inputs) @@ -471,19 +471,16 @@ def _get_video_feat_by_vid(self, vid): def _get_audio_feat_by_vid(self, vid): a_feat_list = [] for _feat_dir in self.a_feat_dirs: - if self.dset_name == 'qvhighlight': - if self.a_feat_types == "clap": - _feat_path = join(_feat_dir, f"{vid}.npz") - _feat = np.load(_feat_path)["features"][:self.max_a_l].astype(np.float32) - elif self.a_feat_types == "pann": + if self.dset_name == 'qvhighlight' or self.dset_name == 'qvhighlight_pretrain': + if self.a_feat_types == "pann": _feat_path = join(_feat_dir, f"{vid}.npy") _feat = np.load(_feat_path)[:self.max_a_l].astype(np.float32) else: - raise NotImplementedError() + raise NotImplementedError _feat = l2_normalize_np_array(_feat) # normalize? a_feat_list.append(_feat) else: - raise NotImplementedError() + raise NotImplementedError # some features are slightly longer than the others min_len = min([len(e) for e in a_feat_list]) diff --git a/training/postprocessing.py b/training/postprocessing.py index 0de21b6..1952ad2 100755 --- a/training/postprocessing.py +++ b/training/postprocessing.py @@ -35,11 +35,7 @@ SOFTWARE. """ -import pprint -import numpy as np import torch -from lighthouse.common.utils.basic_utils import load_jsonl -from training.standalone_eval.eval import eval_submission from tqdm import tqdm